; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X86,X86-NOBMI
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X86,X86-BMINOTBM,X86-BMI1
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X86,X86-BMITBM,X86-BMI1
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X86,X86-BMITBM,X86-BMI2
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X86,X86-BMINOTBM,X86-BMI2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X64,X64-NOBMI
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X64,X64-BMINOTBM,X64-BMI1
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X64,X64-BMITBM,X64-BMI1
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X64,X64-BMITBM,X64-BMI2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=X64,X64-BMINOTBM,X64-BMI2

; *Please* keep in sync with test/CodeGen/AArch64/extract-bits.ll

; https://bugs.llvm.org/show_bug.cgi?id=36419
; https://bugs.llvm.org/show_bug.cgi?id=37603
; https://bugs.llvm.org/show_bug.cgi?id=37610

; Patterns:
;   a) (x >> start) &  (1 << nbits) - 1
;   b) (x >> start) & ~(-1 << nbits)
;   c) (x >> start) &  (-1 >> (32 - y))
;   d) (x >> start) << (32 - y) >> (32 - y)
; are equivalent.

declare void @use32(i32)
declare void @use64(i64)

; ---------------------------------------------------------------------------- ;
; Pattern a. 32-bit
; ---------------------------------------------------------------------------- ;

define i32 @bextr32_a0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_a0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_a0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_a0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_a0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_a0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_a0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_a0_arithmetic(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_a0_arithmetic:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    sarl %cl, %esi
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_a0_arithmetic:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    sarl %cl, %edx
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_a0_arithmetic:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    sarxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_a0_arithmetic:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    sarl %cl, %edi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_a0_arithmetic:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    sarl %cl, %edi
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    bextrl %edx, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_a0_arithmetic:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    sarxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = ashr i32 %val, %numskipbits
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_a1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_a1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_a1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_a1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_a1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_a1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_a1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %conv = zext i8 %numlowbits to i32
  %onebit = shl i32 1, %conv
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_a2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_a2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_a2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, (%eax), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_a2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_a2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %esi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %esi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %esi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_a2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, (%rdi), %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_a2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %shifted = lshr i32 %val, %numskipbits
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_a3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_a3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_a3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, (%eax), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_a3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_a3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %esi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %esi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %esi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_a3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, (%rdi), %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_a3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %conv = zext i8 %numlowbits to i32
  %onebit = shl i32 1, %conv
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_a4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_a4_commutative:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_a4_commutative:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_a4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_a4_commutative:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_a4_commutative:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_a4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %shifted, %mask ; swapped order
  ret i32 %masked
}

define i32 @bextr32_a5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_a5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %ecx
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    movl $1, %esi
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    decl %esi
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %eax, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_a5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl %al, %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl %eax, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $8, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_a5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    bzhil %eax, %edx, %esi
; X86-BMI2-NEXT:    movl %ecx, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    addl $8, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_a5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $1, %ebx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %ebx
; X64-NOBMI-NEXT:    decl %ebx
; X64-NOBMI-NEXT:    andl %edi, %ebx
; X64-NOBMI-NEXT:    movl %esi, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    movl %ebx, %eax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_a5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %ebx
; X64-BMI1-NEXT:    movl %esi, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    movl %ebx, %eax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_a5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %ebx
; X64-BMI2-NEXT:    movl %esi, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %shifted
  call void @use32(i32 %numskipbits)
  ret i32 %masked
}

; 64-bit

define i64 @bextr64_a0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_a0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB7_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB7_2:
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    je .LBB7_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB7_4:
; X86-NOBMI-NEXT:    addl $-1, %eax
; X86-NOBMI-NEXT:    adcl $-1, %edx
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_a0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB7_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB7_2:
; X86-BMI1-NEXT:    movl $1, %eax
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:    movb %ch, %cl
; X86-BMI1-NEXT:    shldl %cl, %eax, %edx
; X86-BMI1-NEXT:    shll %cl, %eax
; X86-BMI1-NEXT:    testb $32, %ch
; X86-BMI1-NEXT:    je .LBB7_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB7_4:
; X86-BMI1-NEXT:    addl $-1, %eax
; X86-BMI1-NEXT:    adcl $-1, %edx
; X86-BMI1-NEXT:    andl %esi, %eax
; X86-BMI1-NEXT:    andl %edi, %edx
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_a0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB7_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB7_2:
; X86-BMI2-NEXT:    movl $1, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    movl %ebx, %ecx
; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB7_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB7_4:
; X86-BMI2-NEXT:    addl $-1, %eax
; X86-BMI2-NEXT:    adcl $-1, %edx
; X86-BMI2-NEXT:    andl %esi, %eax
; X86-BMI2-NEXT:    andl %edi, %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_a0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decq %rax
; X64-NOBMI-NEXT:    andq %rdi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_a0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_a0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %onebit = shl i64 1, %numlowbits
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_a0_arithmetic(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_a0_arithmetic:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:    sarl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %edi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB8_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    sarl $31, %eax
; X86-NOBMI-NEXT:    movl %esi, %edi
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB8_2:
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    je .LBB8_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB8_4:
; X86-NOBMI-NEXT:    addl $-1, %eax
; X86-NOBMI-NEXT:    adcl $-1, %edx
; X86-NOBMI-NEXT:    andl %edi, %eax
; X86-NOBMI-NEXT:    andl %esi, %edx
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_a0_arithmetic:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %eax, %esi
; X86-BMI1-NEXT:    sarl %cl, %esi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %edi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB8_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    sarl $31, %eax
; X86-BMI1-NEXT:    movl %esi, %edi
; X86-BMI1-NEXT:    movl %eax, %esi
; X86-BMI1-NEXT:  .LBB8_2:
; X86-BMI1-NEXT:    movl $1, %eax
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:    movb %ch, %cl
; X86-BMI1-NEXT:    shldl %cl, %eax, %edx
; X86-BMI1-NEXT:    shll %cl, %eax
; X86-BMI1-NEXT:    testb $32, %ch
; X86-BMI1-NEXT:    je .LBB8_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB8_4:
; X86-BMI1-NEXT:    addl $-1, %eax
; X86-BMI1-NEXT:    adcl $-1, %edx
; X86-BMI1-NEXT:    andl %edi, %eax
; X86-BMI1-NEXT:    andl %esi, %edx
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_a0_arithmetic:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    sarxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB8_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    sarl $31, %eax
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    movl %eax, %edi
; X86-BMI2-NEXT:  .LBB8_2:
; X86-BMI2-NEXT:    movl $1, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    movl %ebx, %ecx
; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB8_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB8_4:
; X86-BMI2-NEXT:    addl $-1, %eax
; X86-BMI2-NEXT:    adcl $-1, %edx
; X86-BMI2-NEXT:    andl %esi, %eax
; X86-BMI2-NEXT:    andl %edi, %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_a0_arithmetic:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    sarq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decq %rax
; X64-NOBMI-NEXT:    andq %rdi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_a0_arithmetic:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    movq %rsi, %rcx
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-BMI1-NEXT:    sarq %cl, %rdi
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    bextrq %rdx, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_a0_arithmetic:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    sarxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %shifted = ashr i64 %val, %numskipbits
  %onebit = shl i64 1, %numlowbits
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_a1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_a1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB9_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB9_2:
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    je .LBB9_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB9_4:
; X86-NOBMI-NEXT:    addl $-1, %eax
; X86-NOBMI-NEXT:    adcl $-1, %edx
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_a1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB9_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB9_2:
; X86-BMI1-NEXT:    movl $1, %eax
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:    movb %ch, %cl
; X86-BMI1-NEXT:    shldl %cl, %eax, %edx
; X86-BMI1-NEXT:    shll %cl, %eax
; X86-BMI1-NEXT:    testb $32, %ch
; X86-BMI1-NEXT:    je .LBB9_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB9_4:
; X86-BMI1-NEXT:    addl $-1, %eax
; X86-BMI1-NEXT:    adcl $-1, %edx
; X86-BMI1-NEXT:    andl %esi, %eax
; X86-BMI1-NEXT:    andl %edi, %edx
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_a1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB9_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB9_2:
; X86-BMI2-NEXT:    movl $1, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    movl %ebx, %ecx
; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB9_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB9_4:
; X86-BMI2-NEXT:    addl $-1, %eax
; X86-BMI2-NEXT:    adcl $-1, %edx
; X86-BMI2-NEXT:    andl %esi, %eax
; X86-BMI2-NEXT:    andl %edi, %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_a1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decq %rax
; X64-NOBMI-NEXT:    andq %rdi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_a1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_a1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %conv = zext i8 %numlowbits to i64
  %onebit = shl i64 1, %conv
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_a2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_a2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    movl 4(%eax), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB10_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB10_2:
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    je .LBB10_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB10_4:
; X86-NOBMI-NEXT:    addl $-1, %eax
; X86-NOBMI-NEXT:    adcl $-1, %edx
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_a2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %esi
; X86-BMI1-NEXT:    movl 4(%eax), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB10_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB10_2:
; X86-BMI1-NEXT:    movl $1, %eax
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:    movb %ch, %cl
; X86-BMI1-NEXT:    shldl %cl, %eax, %edx
; X86-BMI1-NEXT:    shll %cl, %eax
; X86-BMI1-NEXT:    testb $32, %ch
; X86-BMI1-NEXT:    je .LBB10_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB10_4:
; X86-BMI1-NEXT:    addl $-1, %eax
; X86-BMI1-NEXT:    adcl $-1, %edx
; X86-BMI1-NEXT:    andl %esi, %eax
; X86-BMI1-NEXT:    andl %edi, %edx
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_a2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl (%eax), %esi
; X86-BMI2-NEXT:    movl 4(%eax), %eax
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB10_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB10_2:
; X86-BMI2-NEXT:    movl $1, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    movl %ebx, %ecx
; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB10_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB10_4:
; X86-BMI2-NEXT:    addl $-1, %eax
; X86-BMI2-NEXT:    adcl $-1, %edx
; X86-BMI2-NEXT:    andl %esi, %eax
; X86-BMI2-NEXT:    andl %edi, %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_a2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq (%rdi), %rsi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rsi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decq %rax
; X64-NOBMI-NEXT:    andq %rsi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_a2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, (%rdi), %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_a2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %shifted = lshr i64 %val, %numskipbits
  %onebit = shl i64 1, %numlowbits
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_a3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_a3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    movl 4(%eax), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB11_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB11_2:
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    je .LBB11_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB11_4:
; X86-NOBMI-NEXT:    addl $-1, %eax
; X86-NOBMI-NEXT:    adcl $-1, %edx
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_a3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %esi
; X86-BMI1-NEXT:    movl 4(%eax), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB11_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB11_2:
; X86-BMI1-NEXT:    movl $1, %eax
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:    movb %ch, %cl
; X86-BMI1-NEXT:    shldl %cl, %eax, %edx
; X86-BMI1-NEXT:    shll %cl, %eax
; X86-BMI1-NEXT:    testb $32, %ch
; X86-BMI1-NEXT:    je .LBB11_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB11_4:
; X86-BMI1-NEXT:    addl $-1, %eax
; X86-BMI1-NEXT:    adcl $-1, %edx
; X86-BMI1-NEXT:    andl %esi, %eax
; X86-BMI1-NEXT:    andl %edi, %edx
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_a3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl (%eax), %esi
; X86-BMI2-NEXT:    movl 4(%eax), %eax
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB11_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB11_2:
; X86-BMI2-NEXT:    movl $1, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    movl %ebx, %ecx
; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB11_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB11_4:
; X86-BMI2-NEXT:    addl $-1, %eax
; X86-BMI2-NEXT:    adcl $-1, %edx
; X86-BMI2-NEXT:    andl %esi, %eax
; X86-BMI2-NEXT:    andl %edi, %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_a3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movq (%rdi), %rsi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %rsi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decq %rax
; X64-NOBMI-NEXT:    andq %rsi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_a3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, (%rdi), %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_a3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %conv = zext i8 %numlowbits to i64
  %onebit = shl i64 1, %conv
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_a4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_a4_commutative:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %edx
; X86-NOBMI-NEXT:    shrl %cl, %edx
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB12_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:  .LBB12_2:
; X86-NOBMI-NEXT:    movl $1, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shldl %cl, %esi, %edi
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    je .LBB12_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %esi, %edi
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:  .LBB12_4:
; X86-NOBMI-NEXT:    addl $-1, %esi
; X86-NOBMI-NEXT:    adcl $-1, %edi
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_a4_commutative:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %esi, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB12_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edx, %eax
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:  .LBB12_2:
; X86-BMI1-NEXT:    movl $1, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:    movb %ch, %cl
; X86-BMI1-NEXT:    shldl %cl, %esi, %edi
; X86-BMI1-NEXT:    shll %cl, %esi
; X86-BMI1-NEXT:    testb $32, %ch
; X86-BMI1-NEXT:    je .LBB12_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %esi, %edi
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:  .LBB12_4:
; X86-BMI1-NEXT:    addl $-1, %esi
; X86-BMI1-NEXT:    adcl $-1, %edi
; X86-BMI1-NEXT:    andl %esi, %eax
; X86-BMI1-NEXT:    andl %edi, %edx
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_a4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB12_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB12_2:
; X86-BMI2-NEXT:    movl $1, %edi
; X86-BMI2-NEXT:    xorl %esi, %esi
; X86-BMI2-NEXT:    movl %ebx, %ecx
; X86-BMI2-NEXT:    shldl %cl, %edi, %esi
; X86-BMI2-NEXT:    shlxl %ebx, %edi, %ecx
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB12_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ecx, %esi
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:  .LBB12_4:
; X86-BMI2-NEXT:    addl $-1, %ecx
; X86-BMI2-NEXT:    adcl $-1, %esi
; X86-BMI2-NEXT:    andl %ecx, %eax
; X86-BMI2-NEXT:    andl %esi, %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_a4_commutative:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decq %rax
; X64-NOBMI-NEXT:    andq %rdi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_a4_commutative:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_a4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %onebit = shl i64 1, %numlowbits
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %shifted, %mask ; swapped order
  ret i64 %masked
}

define i64 @bextr64_a5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_a5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %esi, %ebp
; X86-NOBMI-NEXT:    movl %eax, %ecx
; X86-NOBMI-NEXT:    shrl %cl, %ebp
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %ebx
; X86-NOBMI-NEXT:    testb $32, %al
; X86-NOBMI-NEXT:    je .LBB13_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %ebp, %ebx
; X86-NOBMI-NEXT:    xorl %ebp, %ebp
; X86-NOBMI-NEXT:  .LBB13_2:
; X86-NOBMI-NEXT:    movl $1, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shldl %cl, %esi, %edi
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    testb $32, %dl
; X86-NOBMI-NEXT:    je .LBB13_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %esi, %edi
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:  .LBB13_4:
; X86-NOBMI-NEXT:    addl $-1, %esi
; X86-NOBMI-NEXT:    adcl $-1, %edi
; X86-NOBMI-NEXT:    andl %ebx, %esi
; X86-NOBMI-NEXT:    andl %ebp, %edi
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_a5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %esi, %ebp
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shrl %cl, %ebp
; X86-BMI1-NEXT:    shrdl %cl, %esi, %ebx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB13_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %ebp, %ebx
; X86-BMI1-NEXT:    xorl %ebp, %ebp
; X86-BMI1-NEXT:  .LBB13_2:
; X86-BMI1-NEXT:    movl $1, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:    movl %edx, %ecx
; X86-BMI1-NEXT:    shldl %cl, %esi, %edi
; X86-BMI1-NEXT:    shll %cl, %esi
; X86-BMI1-NEXT:    testb $32, %dl
; X86-BMI1-NEXT:    je .LBB13_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %esi, %edi
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:  .LBB13_4:
; X86-BMI1-NEXT:    addl $-1, %esi
; X86-BMI1-NEXT:    adcl $-1, %edi
; X86-BMI1-NEXT:    andl %ebx, %esi
; X86-BMI1-NEXT:    andl %ebp, %edi
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_a5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl %eax, %ecx
; X86-BMI2-NEXT:    shrdl %cl, %esi, %ebx
; X86-BMI2-NEXT:    shrxl %eax, %esi, %ebp
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB13_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %ebp, %ebx
; X86-BMI2-NEXT:    xorl %ebp, %ebp
; X86-BMI2-NEXT:  .LBB13_2:
; X86-BMI2-NEXT:    movl $1, %edi
; X86-BMI2-NEXT:    xorl %esi, %esi
; X86-BMI2-NEXT:    movl %edx, %ecx
; X86-BMI2-NEXT:    shldl %cl, %edi, %esi
; X86-BMI2-NEXT:    shlxl %edx, %edi, %edi
; X86-BMI2-NEXT:    testb $32, %dl
; X86-BMI2-NEXT:    je .LBB13_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB13_4:
; X86-BMI2-NEXT:    addl $-1, %edi
; X86-BMI2-NEXT:    adcl $-1, %esi
; X86-BMI2-NEXT:    andl %ebx, %edi
; X86-BMI2-NEXT:    andl %ebp, %esi
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    movl %edi, %eax
; X86-BMI2-NEXT:    movl %esi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_a5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %ebx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rbx
; X64-NOBMI-NEXT:    decq %rbx
; X64-NOBMI-NEXT:    andq %rdi, %rbx
; X64-NOBMI-NEXT:    movq %rsi, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_a5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rbx
; X64-BMI1-NEXT:    movq %rsi, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_a5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rbx
; X64-BMI2-NEXT:    movq %rsi, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    movq %rbx, %rax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %onebit = shl i64 1, %numlowbits
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %mask, %shifted
  call void @use64(i64 %numskipbits)
  ret i64 %masked
}

; 64-bit, but with 32-bit output

; Everything done in 64-bit, truncation happens last.
define i32 @bextr64_32_a0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_a0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB14_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB14_2:
; X86-NOBMI-NEXT:    movl $1, %edi
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %edi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:    testb $32, %dl
; X86-NOBMI-NEXT:    jne .LBB14_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %edi, %eax
; X86-NOBMI-NEXT:  .LBB14_4:
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_a0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    shrdl %cl, %edi, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB14_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %esi
; X86-BMI1-NEXT:  .LBB14_2:
; X86-BMI1-NEXT:    movl $1, %edi
; X86-BMI1-NEXT:    movl %edx, %ecx
; X86-BMI1-NEXT:    shll %cl, %edi
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:    testb $32, %dl
; X86-BMI1-NEXT:    jne .LBB14_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %edi, %eax
; X86-BMI1-NEXT:  .LBB14_4:
; X86-BMI1-NEXT:    decl %eax
; X86-BMI1-NEXT:    andl %esi, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_a0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB14_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
; X86-BMI2-NEXT:  .LBB14_2:
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    jne .LBB14_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl $1, %eax
; X86-BMI2-NEXT:    shlxl %ebx, %eax, %eax
; X86-BMI2-NEXT:  .LBB14_4:
; X86-BMI2-NEXT:    decl %eax
; X86-BMI2-NEXT:    andl %edx, %eax
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_a0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_a0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_a0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %onebit = shl i64 1, %numlowbits
  %mask = add nsw i64 %onebit, -1
  %masked = and i64 %mask, %shifted
  %res = trunc i64 %masked to i32
  ret i32 %res
}

; Shifting happens in 64-bit, then truncation. Masking is 32-bit.
define i32 @bextr64_32_a1(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_a1:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB15_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB15_2:
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_a1:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB15_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB15_2:
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_a1:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB15_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB15_2:
; X86-BMI2-NEXT:    bzhil %eax, %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_a1:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_a1:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_a1:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %truncshifted = trunc i64 %shifted to i32
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %truncshifted
  ret i32 %masked
}

; Shifting happens in 64-bit, then truncation (with extra use).
; Masking is 32-bit.
define i32 @bextr64_32_a1_trunc_extrause(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_a1_trunc_extrause:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl %edx, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edx, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB16_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB16_2:
; X86-NOBMI-NEXT:    movl %esi, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %ebx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_a1_trunc_extrause:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB16_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %esi
; X86-BMI1-NEXT:  .LBB16_2:
; X86-BMI1-NEXT:    movl %esi, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    shll $8, %ebx
; X86-BMI1-NEXT:    bextrl %ebx, %esi, %eax
; X86-BMI1-NEXT:    addl $4, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_a1_trunc_extrause:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB16_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %esi
; X86-BMI2-NEXT:  .LBB16_2:
; X86-BMI2-NEXT:    movl %esi, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    bzhil %ebx, %esi, %eax
; X86-BMI2-NEXT:    addl $4, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_a1_trunc_extrause:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbp
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %edx, %ebp
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %rbx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movl %ebx, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %ebp, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %ebx, %eax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %rbp
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_a1_trunc_extrause:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbp
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %edx, %ebp
; X64-BMI1-NEXT:    movq %rsi, %rcx
; X64-BMI1-NEXT:    movq %rdi, %rbx
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-BMI1-NEXT:    shrq %cl, %rbx
; X64-BMI1-NEXT:    movl %ebx, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    shll $8, %ebp
; X64-BMI1-NEXT:    bextrl %ebp, %ebx, %eax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %rbp
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_a1_trunc_extrause:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbp
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebp
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rbx
; X64-BMI2-NEXT:    movl %ebx, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    bzhil %ebp, %ebx, %eax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %rbp
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %truncshifted = trunc i64 %shifted to i32
  call void @use32(i32 %truncshifted)
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %masked = and i32 %mask, %truncshifted
  ret i32 %masked
}

; Shifting happens in 64-bit. Mask is 32-bit, but extended to 64-bit.
; Masking is 64-bit. Then truncation.
define i32 @bextr64_32_a2(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_a2:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB17_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB17_2:
; X86-NOBMI-NEXT:    movl $1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_a2:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB17_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB17_2:
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_a2:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB17_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB17_2:
; X86-BMI2-NEXT:    bzhil %eax, %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_a2:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_a2:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_a2:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %onebit = shl i32 1, %numlowbits
  %mask = add nsw i32 %onebit, -1
  %zextmask = zext i32 %mask to i64
  %masked = and i64 %zextmask, %shifted
  %truncmasked = trunc i64 %masked to i32
  ret i32 %truncmasked
}

; Shifting happens in 64-bit. Mask is 32-bit, but calculated in 64-bit.
; Masking is 64-bit. Then truncation.
define i32 @bextr64_32_a3(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_a3:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB18_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB18_2:
; X86-NOBMI-NEXT:    movl $1, %edi
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %edi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:    testb $32, %dl
; X86-NOBMI-NEXT:    jne .LBB18_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %edi, %eax
; X86-NOBMI-NEXT:  .LBB18_4:
; X86-NOBMI-NEXT:    decl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_a3:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    shrdl %cl, %edi, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB18_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %esi
; X86-BMI1-NEXT:  .LBB18_2:
; X86-BMI1-NEXT:    movl $1, %edi
; X86-BMI1-NEXT:    movl %edx, %ecx
; X86-BMI1-NEXT:    shll %cl, %edi
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:    testb $32, %dl
; X86-BMI1-NEXT:    jne .LBB18_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %edi, %eax
; X86-BMI1-NEXT:  .LBB18_4:
; X86-BMI1-NEXT:    decl %eax
; X86-BMI1-NEXT:    andl %esi, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_a3:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB18_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
; X86-BMI2-NEXT:  .LBB18_2:
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    jne .LBB18_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl $1, %eax
; X86-BMI2-NEXT:    shlxl %ebx, %eax, %eax
; X86-BMI2-NEXT:  .LBB18_4:
; X86-BMI2-NEXT:    decl %eax
; X86-BMI2-NEXT:    andl %edx, %eax
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_a3:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    decl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_a3:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_a3:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %onebit = shl i64 1, %numlowbits
  %mask = add nsw i64 %onebit, 4294967295
  %masked = and i64 %mask, %shifted
  %truncmasked = trunc i64 %masked to i32
  ret i32 %truncmasked
}

; ---------------------------------------------------------------------------- ;
; Pattern b. 32-bit
; ---------------------------------------------------------------------------- ;

define i32 @bextr32_b0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_b0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_b0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_b0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_b0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $-1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_b0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_b0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %notmask = shl i32 -1, %numlowbits
  %mask = xor i32 %notmask, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_b1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_b1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_b1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_b1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_b1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $-1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_b1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_b1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %conv = zext i8 %numlowbits to i32
  %notmask = shl i32 -1, %conv
  %mask = xor i32 %notmask, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_b2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_b2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_b2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, (%eax), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_b2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_b2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %esi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %esi
; X64-NOBMI-NEXT:    movl $-1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %esi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_b2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, (%rdi), %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_b2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %shifted = lshr i32 %val, %numskipbits
  %notmask = shl i32 -1, %numlowbits
  %mask = xor i32 %notmask, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_b3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_b3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_b3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, (%eax), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_b3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_b3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %esi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %esi
; X64-NOBMI-NEXT:    movl $-1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %esi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_b3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, (%rdi), %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_b3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %conv = zext i8 %numlowbits to i32
  %notmask = shl i32 -1, %conv
  %mask = xor i32 %notmask, -1
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_b4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_b4_commutative:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_b4_commutative:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_b4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_b4_commutative:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $-1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_b4_commutative:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_b4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %notmask = shl i32 -1, %numlowbits
  %mask = xor i32 %notmask, -1
  %masked = and i32 %shifted, %mask ; swapped order
  ret i32 %masked
}

define i32 @bextr32_b5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_b5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %ecx
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    notl %esi
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %eax, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_b5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl %al, %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl %eax, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $8, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_b5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    bzhil %eax, %edx, %esi
; X86-BMI2-NEXT:    movl %ecx, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    addl $8, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_b5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %edi
; X64-NOBMI-NEXT:    movl $-1, %ebx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %ebx
; X64-NOBMI-NEXT:    notl %ebx
; X64-NOBMI-NEXT:    andl %edi, %ebx
; X64-NOBMI-NEXT:    movl %esi, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    movl %ebx, %eax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_b5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %ebx
; X64-BMI1-NEXT:    movl %esi, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    movl %ebx, %eax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_b5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %ebx
; X64-BMI2-NEXT:    movl %esi, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %notmask = shl i32 -1, %numlowbits
  %mask = xor i32 %notmask, -1
  %masked = and i32 %mask, %shifted
  call void @use32(i32 %numskipbits)
  ret i32 %masked
}

; 64-bit

define i64 @bextr64_b0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_b0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB25_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB25_2:
; X86-NOBMI-NEXT:    movl $-1, %edx
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shll %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    jne .LBB25_3
; X86-NOBMI-NEXT:  # %bb.4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    jmp .LBB25_5
; X86-NOBMI-NEXT:  .LBB25_3:
; X86-NOBMI-NEXT:    movl %ebx, %edx
; X86-NOBMI-NEXT:  .LBB25_5:
; X86-NOBMI-NEXT:    notl %edx
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_b0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB25_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:  .LBB25_2:
; X86-BMI1-NEXT:    movl $-1, %edi
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shll %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB25_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %edi
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB25_4:
; X86-BMI1-NEXT:    andnl %edx, %edi, %edx
; X86-BMI1-NEXT:    andnl %esi, %ebx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_b0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB25_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB25_2:
; X86-BMI2-NEXT:    movl $-1, %esi
; X86-BMI2-NEXT:    shlxl %ebx, %esi, %ecx
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB25_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ecx, %esi
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:  .LBB25_4:
; X86-BMI2-NEXT:    andnl %edx, %esi, %edx
; X86-BMI2-NEXT:    andnl %eax, %ecx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_b0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movq $-1, %rax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    notq %rax
; X64-NOBMI-NEXT:    andq %rdi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_b0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_b0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %notmask = shl i64 -1, %numlowbits
  %mask = xor i64 %notmask, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_b1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_b1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB26_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB26_2:
; X86-NOBMI-NEXT:    movl $-1, %edx
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shll %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    jne .LBB26_3
; X86-NOBMI-NEXT:  # %bb.4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    jmp .LBB26_5
; X86-NOBMI-NEXT:  .LBB26_3:
; X86-NOBMI-NEXT:    movl %ebx, %edx
; X86-NOBMI-NEXT:  .LBB26_5:
; X86-NOBMI-NEXT:    notl %edx
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_b1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB26_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:  .LBB26_2:
; X86-BMI1-NEXT:    movl $-1, %edi
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shll %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB26_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %edi
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB26_4:
; X86-BMI1-NEXT:    andnl %edx, %edi, %edx
; X86-BMI1-NEXT:    andnl %esi, %ebx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_b1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB26_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB26_2:
; X86-BMI2-NEXT:    movl $-1, %esi
; X86-BMI2-NEXT:    shlxl %ebx, %esi, %ecx
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB26_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ecx, %esi
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:  .LBB26_4:
; X86-BMI2-NEXT:    andnl %edx, %esi, %edx
; X86-BMI2-NEXT:    andnl %eax, %ecx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_b1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movq $-1, %rax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    notq %rax
; X64-NOBMI-NEXT:    andq %rdi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_b1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_b1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %conv = zext i8 %numlowbits to i64
  %notmask = shl i64 -1, %conv
  %mask = xor i64 %notmask, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_b2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_b2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    movl 4(%eax), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB27_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB27_2:
; X86-NOBMI-NEXT:    movl $-1, %edx
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shll %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    jne .LBB27_3
; X86-NOBMI-NEXT:  # %bb.4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    jmp .LBB27_5
; X86-NOBMI-NEXT:  .LBB27_3:
; X86-NOBMI-NEXT:    movl %ebx, %edx
; X86-NOBMI-NEXT:  .LBB27_5:
; X86-NOBMI-NEXT:    notl %edx
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_b2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    movl (%edx), %esi
; X86-BMI1-NEXT:    movl 4(%edx), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB27_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:  .LBB27_2:
; X86-BMI1-NEXT:    movl $-1, %edi
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shll %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB27_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %edi
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB27_4:
; X86-BMI1-NEXT:    andnl %edx, %edi, %edx
; X86-BMI1-NEXT:    andnl %esi, %ebx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_b2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl (%edx), %eax
; X86-BMI2-NEXT:    movl 4(%edx), %esi
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:    shrdl %cl, %esi, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB27_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB27_2:
; X86-BMI2-NEXT:    movl $-1, %esi
; X86-BMI2-NEXT:    shlxl %ebx, %esi, %ecx
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB27_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ecx, %esi
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:  .LBB27_4:
; X86-BMI2-NEXT:    andnl %edx, %esi, %edx
; X86-BMI2-NEXT:    andnl %eax, %ecx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_b2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq (%rdi), %rsi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rsi
; X64-NOBMI-NEXT:    movq $-1, %rax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    notq %rax
; X64-NOBMI-NEXT:    andq %rsi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_b2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, (%rdi), %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_b2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %shifted = lshr i64 %val, %numskipbits
  %notmask = shl i64 -1, %numlowbits
  %mask = xor i64 %notmask, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_b3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_b3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    movl 4(%eax), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB28_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB28_2:
; X86-NOBMI-NEXT:    movl $-1, %edx
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shll %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    jne .LBB28_3
; X86-NOBMI-NEXT:  # %bb.4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    jmp .LBB28_5
; X86-NOBMI-NEXT:  .LBB28_3:
; X86-NOBMI-NEXT:    movl %ebx, %edx
; X86-NOBMI-NEXT:  .LBB28_5:
; X86-NOBMI-NEXT:    notl %edx
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_b3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    movl (%edx), %esi
; X86-BMI1-NEXT:    movl 4(%edx), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB28_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:  .LBB28_2:
; X86-BMI1-NEXT:    movl $-1, %edi
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shll %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB28_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %edi
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB28_4:
; X86-BMI1-NEXT:    andnl %edx, %edi, %edx
; X86-BMI1-NEXT:    andnl %esi, %ebx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_b3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl (%edx), %eax
; X86-BMI2-NEXT:    movl 4(%edx), %esi
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:    shrdl %cl, %esi, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB28_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB28_2:
; X86-BMI2-NEXT:    movl $-1, %esi
; X86-BMI2-NEXT:    shlxl %ebx, %esi, %ecx
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB28_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ecx, %esi
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:  .LBB28_4:
; X86-BMI2-NEXT:    andnl %edx, %esi, %edx
; X86-BMI2-NEXT:    andnl %eax, %ecx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_b3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movq (%rdi), %rsi
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %rsi
; X64-NOBMI-NEXT:    movq $-1, %rax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    notq %rax
; X64-NOBMI-NEXT:    andq %rsi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_b3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, (%rdi), %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_b3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %conv = zext i8 %numlowbits to i64
  %notmask = shl i64 -1, %conv
  %mask = xor i64 %notmask, -1
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_b4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_b4_commutative:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %edx
; X86-NOBMI-NEXT:    shrl %cl, %edx
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %eax
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB29_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:  .LBB29_2:
; X86-NOBMI-NEXT:    movl $-1, %edi
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shll %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    jne .LBB29_3
; X86-NOBMI-NEXT:  # %bb.4:
; X86-NOBMI-NEXT:    movl %ebx, %esi
; X86-NOBMI-NEXT:    jmp .LBB29_5
; X86-NOBMI-NEXT:  .LBB29_3:
; X86-NOBMI-NEXT:    movl %ebx, %edi
; X86-NOBMI-NEXT:  .LBB29_5:
; X86-NOBMI-NEXT:    notl %edi
; X86-NOBMI-NEXT:    andl %edi, %edx
; X86-NOBMI-NEXT:    notl %esi
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_b4_commutative:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB29_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:  .LBB29_2:
; X86-BMI1-NEXT:    movl $-1, %edi
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shll %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB29_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %edi
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB29_4:
; X86-BMI1-NEXT:    andnl %edx, %edi, %edx
; X86-BMI1-NEXT:    andnl %esi, %ebx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_b4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB29_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB29_2:
; X86-BMI2-NEXT:    movl $-1, %esi
; X86-BMI2-NEXT:    shlxl %ebx, %esi, %ecx
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB29_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ecx, %esi
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:  .LBB29_4:
; X86-BMI2-NEXT:    andnl %edx, %esi, %edx
; X86-BMI2-NEXT:    andnl %eax, %ecx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_b4_commutative:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movq $-1, %rax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    notq %rax
; X64-NOBMI-NEXT:    andq %rdi, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_b4_commutative:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_b4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %notmask = shl i64 -1, %numlowbits
  %mask = xor i64 %notmask, -1
  %masked = and i64 %shifted, %mask ; swapped order
  ret i64 %masked
}

define i64 @bextr64_b5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_b5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %esi, %ebp
; X86-NOBMI-NEXT:    movb %al, %cl
; X86-NOBMI-NEXT:    shrl %cl, %ebp
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %edx
; X86-NOBMI-NEXT:    xorl %ebx, %ebx
; X86-NOBMI-NEXT:    testb $32, %al
; X86-NOBMI-NEXT:    je .LBB30_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %ebp, %edx
; X86-NOBMI-NEXT:    xorl %ebp, %ebp
; X86-NOBMI-NEXT:  .LBB30_2:
; X86-NOBMI-NEXT:    movl $-1, %edi
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    movb %ch, %cl
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    testb $32, %ch
; X86-NOBMI-NEXT:    jne .LBB30_3
; X86-NOBMI-NEXT:  # %bb.4:
; X86-NOBMI-NEXT:    movl %esi, %ebx
; X86-NOBMI-NEXT:    jmp .LBB30_5
; X86-NOBMI-NEXT:  .LBB30_3:
; X86-NOBMI-NEXT:    movl %esi, %edi
; X86-NOBMI-NEXT:  .LBB30_5:
; X86-NOBMI-NEXT:    notl %edi
; X86-NOBMI-NEXT:    andl %ebp, %edi
; X86-NOBMI-NEXT:    notl %ebx
; X86-NOBMI-NEXT:    andl %edx, %ebx
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_b5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %ebx, %esi
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    shrdl %cl, %ebx, %edi
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB30_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edi
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:  .LBB30_2:
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    movl $-1, %ebp
; X86-BMI1-NEXT:    movl %edx, %ecx
; X86-BMI1-NEXT:    shll %cl, %ebp
; X86-BMI1-NEXT:    testb $32, %dl
; X86-BMI1-NEXT:    je .LBB30_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebp, %ebx
; X86-BMI1-NEXT:    xorl %ebp, %ebp
; X86-BMI1-NEXT:  .LBB30_4:
; X86-BMI1-NEXT:    andnl %esi, %ebx, %esi
; X86-BMI1-NEXT:    andnl %edi, %ebp, %edi
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    movl %edi, %eax
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_b5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB30_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB30_2:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
; X86-BMI2-NEXT:    movl $-1, %esi
; X86-BMI2-NEXT:    shlxl %ebx, %esi, %edi
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB30_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB30_4:
; X86-BMI2-NEXT:    andnl %edx, %esi, %esi
; X86-BMI2-NEXT:    andnl %eax, %edi, %edi
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ecx
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    movl %edi, %eax
; X86-BMI2-NEXT:    movl %esi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_b5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movq $-1, %rbx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rbx
; X64-NOBMI-NEXT:    notq %rbx
; X64-NOBMI-NEXT:    andq %rdi, %rbx
; X64-NOBMI-NEXT:    movq %rsi, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_b5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rbx
; X64-BMI1-NEXT:    movq %rsi, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_b5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rbx
; X64-BMI2-NEXT:    movq %rsi, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    movq %rbx, %rax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %notmask = shl i64 -1, %numlowbits
  %mask = xor i64 %notmask, -1
  %masked = and i64 %mask, %shifted
  call void @use64(i64 %numskipbits)
  ret i64 %masked
}

; 64-bit, but with 32-bit output

; Everything done in 64-bit, truncation happens last.
define i32 @bextr64_32_b0(i64 %val, i64 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_b0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB31_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:  .LBB31_2:
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    testb $32, %dl
; X86-NOBMI-NEXT:    jne .LBB31_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %esi, %ecx
; X86-NOBMI-NEXT:  .LBB31_4:
; X86-NOBMI-NEXT:    notl %ecx
; X86-NOBMI-NEXT:    andl %ecx, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_b0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB31_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB31_2:
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shll %cl, %esi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    jne .LBB31_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %esi, %ecx
; X86-BMI1-NEXT:  .LBB31_4:
; X86-BMI1-NEXT:    andnl %edx, %ecx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_b0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB31_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB31_2:
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    jne .LBB31_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %ecx
; X86-BMI2-NEXT:  .LBB31_4:
; X86-BMI2-NEXT:    andnl %edx, %ecx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_b0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movq $-1, %rax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_b0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_b0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shiftedval = lshr i64 %val, %numskipbits
  %widenumlowbits = zext i8 %numlowbits to i64
  %notmask = shl nsw i64 -1, %widenumlowbits
  %mask = xor i64 %notmask, -1
  %wideres = and i64 %shiftedval, %mask
  %res = trunc i64 %wideres to i32
  ret i32 %res
}

; Shifting happens in 64-bit, then truncation. Masking is 32-bit.
define i32 @bextr64_32_b1(i64 %val, i64 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_b1:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB32_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB32_2:
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_b1:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB32_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB32_2:
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_b1:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB32_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB32_2:
; X86-BMI2-NEXT:    bzhil %eax, %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_b1:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $-1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_b1:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_b1:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shiftedval = lshr i64 %val, %numskipbits
  %truncshiftedval = trunc i64 %shiftedval to i32
  %widenumlowbits = zext i8 %numlowbits to i32
  %notmask = shl nsw i32 -1, %widenumlowbits
  %mask = xor i32 %notmask, -1
  %res = and i32 %truncshiftedval, %mask
  ret i32 %res
}

; Shifting happens in 64-bit. Mask is 32-bit, but extended to 64-bit.
; Masking is 64-bit. Then truncation.
define i32 @bextr64_32_b2(i64 %val, i64 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_b2:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB33_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %esi
; X86-NOBMI-NEXT:  .LBB33_2:
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    notl %eax
; X86-NOBMI-NEXT:    andl %esi, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_b2:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB33_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB33_2:
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_b2:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB33_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB33_2:
; X86-BMI2-NEXT:    bzhil %eax, %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_b2:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $-1, %eax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    notl %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_b2:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_b2:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shiftedval = lshr i64 %val, %numskipbits
  %widenumlowbits = zext i8 %numlowbits to i32
  %notmask = shl nsw i32 -1, %widenumlowbits
  %mask = xor i32 %notmask, -1
  %zextmask = zext i32 %mask to i64
  %wideres = and i64 %shiftedval, %zextmask
  %res = trunc i64 %wideres to i32
  ret i32 %res
}

; Shifting happens in 64-bit. Mask is 32-bit, but calculated in 64-bit.
; Masking is 64-bit. Then truncation.
define i32 @bextr64_32_b3(i64 %val, i64 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_b3:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl %edi, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %edi, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB34_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:  .LBB34_2:
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    movl %edx, %ecx
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    testb $32, %dl
; X86-NOBMI-NEXT:    jne .LBB34_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %esi, %ecx
; X86-NOBMI-NEXT:  .LBB34_4:
; X86-NOBMI-NEXT:    notl %ecx
; X86-NOBMI-NEXT:    andl %ecx, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_b3:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB34_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB34_2:
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shll %cl, %esi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    jne .LBB34_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %esi, %ecx
; X86-BMI1-NEXT:  .LBB34_4:
; X86-BMI1-NEXT:    andnl %edx, %ecx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_b3:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB34_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB34_2:
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    jne .LBB34_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %ecx
; X86-BMI2-NEXT:  .LBB34_4:
; X86-BMI2-NEXT:    andnl %edx, %ecx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_b3:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-NOBMI-NEXT:    movl $4294967295, %esi # imm = 0xFFFFFFFF
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rsi
; X64-NOBMI-NEXT:    xorl %esi, %eax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_b3:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_b3:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shiftedval = lshr i64 %val, %numskipbits
  %widenumlowbits = zext i8 %numlowbits to i64
  %notmask = shl nsw i64 4294967295, %widenumlowbits
  %mask = xor i64 %notmask, 4294967295
  %wideres = and i64 %shiftedval, %mask
  %res = trunc i64 %wideres to i32
  ret i32 %res
}

; ---------------------------------------------------------------------------- ;
; Pattern c. 32-bit
; ---------------------------------------------------------------------------- ;

define i32 @bextr32_c0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_c0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl %esi, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_c0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    movl %esi, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    andl %edi, %esi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $4, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_c0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl %ebx, %eax
; X86-BMI2-NEXT:    negb %al
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    movl %eax, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    bzhil %ebx, %esi, %eax
; X86-BMI2-NEXT:    addl $4, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_c0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbp
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl %edi, %ebx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl $-1, %ebp
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebp
; X64-NOBMI-NEXT:    movl %ebp, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    andl %ebx, %ebp
; X64-NOBMI-NEXT:    movl %ebp, %eax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %rbp
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_c0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbp
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    movl %edi, %ebx
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    shrl %cl, %ebx
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movl $-1, %ebp
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrl %cl, %ebp
; X64-BMI1-NEXT:    movl %ebp, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    andl %ebx, %ebp
; X64-BMI1-NEXT:    movl %ebp, %eax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %rbp
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_c0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbp
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    shrxl %esi, %edi, %ebp
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movl $-1, %ecx
; X64-BMI2-NEXT:    shrxl %eax, %ecx, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    bzhil %ebx, %ebp, %eax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %rbp
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %mask = lshr i32 -1, %numhighbits
  call void @use32(i32 %mask)
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_c1_indexzext(i32 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_c1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl %esi, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_c1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    movl %esi, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    andl %edi, %esi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $4, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_c1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl %ebx, %eax
; X86-BMI2-NEXT:    negb %al
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    movl %eax, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    bzhil %ebx, %esi, %eax
; X86-BMI2-NEXT:    addl $4, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_c1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbp
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl %edi, %ebx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl $-1, %ebp
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebp
; X64-NOBMI-NEXT:    movl %ebp, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    andl %ebx, %ebp
; X64-NOBMI-NEXT:    movl %ebp, %eax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %rbp
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_c1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbp
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    movl %edi, %ebx
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    shrl %cl, %ebx
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movl $-1, %ebp
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrl %cl, %ebp
; X64-BMI1-NEXT:    movl %ebp, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    andl %ebx, %ebp
; X64-BMI1-NEXT:    movl %ebp, %eax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %rbp
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_c1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbp
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    shrxl %esi, %edi, %ebp
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movl $-1, %ecx
; X64-BMI2-NEXT:    shrxl %eax, %ecx, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    bzhil %ebx, %ebp, %eax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %rbp
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %numhighbits = sub i8 32, %numlowbits
  %sh_prom = zext i8 %numhighbits to i32
  %mask = lshr i32 -1, %sh_prom
  call void @use32(i32 %mask)
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_c2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_c2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl %esi, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_c2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    movl %esi, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    andl %edi, %esi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $4, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_c2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, (%eax), %esi
; X86-BMI2-NEXT:    movl %ebx, %eax
; X86-BMI2-NEXT:    negb %al
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    movl %eax, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    bzhil %ebx, %esi, %eax
; X86-BMI2-NEXT:    addl $4, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_c2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbp
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %ebp
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebp
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl $-1, %ebx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    movl %ebx, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    andl %ebp, %ebx
; X64-NOBMI-NEXT:    movl %ebx, %eax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %rbp
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_c2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbp
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    movl (%rdi), %ebp
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    shrl %cl, %ebp
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movl $-1, %ebx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrl %cl, %ebx
; X64-BMI1-NEXT:    movl %ebx, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    andl %ebp, %ebx
; X64-BMI1-NEXT:    movl %ebx, %eax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %rbp
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_c2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbp
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %ebp
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movl $-1, %ecx
; X64-BMI2-NEXT:    shrxl %eax, %ecx, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    bzhil %ebx, %ebp, %eax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %rbp
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %shifted = lshr i32 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %mask = lshr i32 -1, %numhighbits
  call void @use32(i32 %mask)
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_c3_load_indexzext(i32* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_c3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl %esi, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_c3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    movl %esi, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    andl %edi, %esi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $4, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_c3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, (%eax), %esi
; X86-BMI2-NEXT:    movl %ebx, %eax
; X86-BMI2-NEXT:    negb %al
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    movl %eax, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    bzhil %ebx, %esi, %eax
; X86-BMI2-NEXT:    addl $4, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_c3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbp
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %ebp
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebp
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl $-1, %ebx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    movl %ebx, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    andl %ebp, %ebx
; X64-NOBMI-NEXT:    movl %ebx, %eax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %rbp
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_c3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbp
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    movl (%rdi), %ebp
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    shrl %cl, %ebp
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movl $-1, %ebx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrl %cl, %ebx
; X64-BMI1-NEXT:    movl %ebx, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    andl %ebp, %ebx
; X64-BMI1-NEXT:    movl %ebx, %eax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %rbp
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_c3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbp
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %ebp
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movl $-1, %ecx
; X64-BMI2-NEXT:    shrxl %eax, %ecx, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    bzhil %ebx, %ebp, %eax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %rbp
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %numhighbits = sub i8 32, %numlowbits
  %sh_prom = zext i8 %numhighbits to i32
  %mask = lshr i32 -1, %sh_prom
  call void @use32(i32 %mask)
  %masked = and i32 %mask, %shifted
  ret i32 %masked
}

define i32 @bextr32_c4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_c4_commutative:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl %esi, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $4, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_c4_commutative:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    movl %esi, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    andl %edi, %esi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $4, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_c4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl %ebx, %eax
; X86-BMI2-NEXT:    negb %al
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    movl %eax, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    bzhil %ebx, %esi, %eax
; X86-BMI2-NEXT:    addl $4, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_c4_commutative:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbp
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl %edi, %ebx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl $-1, %ebp
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebp
; X64-NOBMI-NEXT:    movl %ebp, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    andl %ebx, %ebp
; X64-NOBMI-NEXT:    movl %ebp, %eax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %rbp
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_c4_commutative:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbp
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    movl %edi, %ebx
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    shrl %cl, %ebx
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movl $-1, %ebp
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrl %cl, %ebp
; X64-BMI1-NEXT:    movl %ebp, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    andl %ebx, %ebp
; X64-BMI1-NEXT:    movl %ebp, %eax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %rbp
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_c4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbp
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    shrxl %esi, %edi, %ebp
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movl $-1, %ecx
; X64-BMI2-NEXT:    shrxl %eax, %ecx, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    bzhil %ebx, %ebp, %eax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %rbp
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %mask = lshr i32 -1, %numhighbits
  call void @use32(i32 %mask)
  %masked = and i32 %shifted, %mask ; swapped order
  ret i32 %masked
}

define i32 @bextr32_c5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_c5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $16, %esp
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-NOBMI-NEXT:    movl %ebx, %ecx
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl %esi, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    andl %edi, %esi
; X86-NOBMI-NEXT:    movl %ebx, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_c5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $16, %esp
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-BMI1-NEXT:    movl %ebx, %ecx
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    xorl %ecx, %ecx
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    movl %esi, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    andl %edi, %esi
; X86-BMI1-NEXT:    movl %ebx, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_c5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $16, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI2-NEXT:    shrxl %edi, {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl %ebx, %eax
; X86-BMI2-NEXT:    negb %al
; X86-BMI2-NEXT:    movl $-1, %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    movl %eax, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    bzhil %ebx, %esi, %esi
; X86-BMI2-NEXT:    movl %edi, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_c5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbp
; X64-NOBMI-NEXT:    pushq %r14
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movl %esi, %r14d
; X64-NOBMI-NEXT:    movl %edi, %ebp
; X64-NOBMI-NEXT:    movl %r14d, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebp
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl $-1, %ebx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    movl %ebx, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    andl %ebp, %ebx
; X64-NOBMI-NEXT:    movl %r14d, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    movl %ebx, %eax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %r14
; X64-NOBMI-NEXT:    popq %rbp
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_c5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbp
; X64-BMI1-NEXT:    pushq %r14
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    movl %esi, %r14d
; X64-BMI1-NEXT:    movl %edi, %ebp
; X64-BMI1-NEXT:    movl %r14d, %ecx
; X64-BMI1-NEXT:    shrl %cl, %ebp
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movl $-1, %ebx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrl %cl, %ebx
; X64-BMI1-NEXT:    movl %ebx, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    andl %ebp, %ebx
; X64-BMI1-NEXT:    movl %r14d, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    movl %ebx, %eax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %r14
; X64-BMI1-NEXT:    popq %rbp
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_c5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbp
; X64-BMI2-NEXT:    pushq %r14
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    movl %esi, %ebp
; X64-BMI2-NEXT:    shrxl %esi, %edi, %r14d
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movl $-1, %ecx
; X64-BMI2-NEXT:    shrxl %eax, %ecx, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    bzhil %ebx, %r14d, %ebx
; X64-BMI2-NEXT:    movl %ebp, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %r14
; X64-BMI2-NEXT:    popq %rbp
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %mask = lshr i32 -1, %numhighbits
  call void @use32(i32 %mask)
  %masked = and i32 %mask, %shifted
  call void @use32(i32 %numskipbits)
  ret i32 %masked
}

; 64-bit

define i64 @bextr64_c0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_c0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB41_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB41_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %ebp
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    shrl %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB41_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %ebx, %ebp
; X86-NOBMI-NEXT:    xorl %ebx, %ebx
; X86-NOBMI-NEXT:  .LBB41_4:
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    andl %ebp, %esi
; X86-NOBMI-NEXT:    andl %ebx, %edi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_c0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB41_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB41_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %ebp
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    shrl %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB41_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %ebp
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB41_4:
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    andl %ebp, %esi
; X86-BMI1-NEXT:    andl %ebx, %edi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_c0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB41_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB41_2:
; X86-BMI2-NEXT:    movb $64, %al
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl $-1, %ebp
; X86-BMI2-NEXT:    shrxl %eax, %ebp, %ebx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB41_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ebx, %ebp
; X86-BMI2-NEXT:    xorl %ebx, %ebx
; X86-BMI2-NEXT:  .LBB41_4:
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    andl %ebp, %esi
; X86-BMI2-NEXT:    andl %ebx, %edi
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    movl %edi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_c0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %r14
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %r14
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %r14
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movq $-1, %rbx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    andq %r14, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %r14
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_c0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %r14
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movq %rsi, %rcx
; X64-BMI1-NEXT:    movq %rdi, %r14
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-BMI1-NEXT:    shrq %cl, %r14
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movq $-1, %rbx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrq %cl, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    andq %r14, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %r14
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_c0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %r14
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movq %rdx, %rbx
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %r14
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movq $-1, %rcx
; X64-BMI2-NEXT:    shrxq %rax, %rcx, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    bzhiq %rbx, %r14, %rax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %r14
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %mask = lshr i64 -1, %numhighbits
  call void @use64(i64 %mask)
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_c1_indexzext(i64 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_c1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB42_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB42_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %ebp
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    shrl %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB42_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %ebx, %ebp
; X86-NOBMI-NEXT:    xorl %ebx, %ebx
; X86-NOBMI-NEXT:  .LBB42_4:
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    andl %ebp, %esi
; X86-NOBMI-NEXT:    andl %ebx, %edi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_c1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB42_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB42_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %ebp
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    shrl %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB42_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %ebp
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB42_4:
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    andl %ebp, %esi
; X86-BMI1-NEXT:    andl %ebx, %edi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_c1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB42_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB42_2:
; X86-BMI2-NEXT:    movb $64, %al
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl $-1, %ebp
; X86-BMI2-NEXT:    shrxl %eax, %ebp, %ebx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB42_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ebx, %ebp
; X86-BMI2-NEXT:    xorl %ebx, %ebx
; X86-BMI2-NEXT:  .LBB42_4:
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    andl %ebp, %esi
; X86-BMI2-NEXT:    andl %ebx, %edi
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    movl %edi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_c1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %r14
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movq %rdi, %r14
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %r14
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movq $-1, %rbx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    andq %r14, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %r14
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_c1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %r14
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    movq %rdi, %r14
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    shrq %cl, %r14
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movq $-1, %rbx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrq %cl, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    andq %r14, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %r14
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_c1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %r14
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %r14
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movq $-1, %rcx
; X64-BMI2-NEXT:    shrxq %rax, %rcx, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    bzhiq %rbx, %r14, %rax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %r14
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %numhighbits = sub i8 64, %numlowbits
  %sh_prom = zext i8 %numhighbits to i64
  %mask = lshr i64 -1, %sh_prom
  call void @use64(i64 %mask)
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_c2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_c2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    movl 4(%eax), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB43_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB43_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %ebp
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    shrl %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB43_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %ebx, %ebp
; X86-NOBMI-NEXT:    xorl %ebx, %ebx
; X86-NOBMI-NEXT:  .LBB43_4:
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    andl %ebp, %esi
; X86-NOBMI-NEXT:    andl %ebx, %edi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_c2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %esi
; X86-BMI1-NEXT:    movl 4(%eax), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB43_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB43_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %ebp
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    shrl %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB43_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %ebp
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB43_4:
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    andl %ebp, %esi
; X86-BMI1-NEXT:    andl %ebx, %edi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_c2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl (%eax), %esi
; X86-BMI2-NEXT:    movl 4(%eax), %eax
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB43_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB43_2:
; X86-BMI2-NEXT:    movb $64, %al
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl $-1, %ebp
; X86-BMI2-NEXT:    shrxl %eax, %ebp, %ebx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB43_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ebx, %ebp
; X86-BMI2-NEXT:    xorl %ebx, %ebx
; X86-BMI2-NEXT:  .LBB43_4:
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    andl %ebp, %esi
; X86-BMI2-NEXT:    andl %ebx, %edi
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    movl %edi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_c2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %r14
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq (%rdi), %r14
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %r14
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movq $-1, %rbx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    andq %r14, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %r14
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_c2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %r14
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movq %rsi, %rcx
; X64-BMI1-NEXT:    movq (%rdi), %r14
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-BMI1-NEXT:    shrq %cl, %r14
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movq $-1, %rbx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrq %cl, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    andq %r14, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %r14
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_c2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %r14
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movq %rdx, %rbx
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %r14
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movq $-1, %rcx
; X64-BMI2-NEXT:    shrxq %rax, %rcx, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    bzhiq %rbx, %r14, %rax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %r14
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %mask = lshr i64 -1, %numhighbits
  call void @use64(i64 %mask)
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_c3_load_indexzext(i64* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_c3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %esi
; X86-NOBMI-NEXT:    movl 4(%eax), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB44_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB44_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %ebp
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    shrl %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB44_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %ebx, %ebp
; X86-NOBMI-NEXT:    xorl %ebx, %ebx
; X86-NOBMI-NEXT:  .LBB44_4:
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    andl %ebp, %esi
; X86-NOBMI-NEXT:    andl %ebx, %edi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_c3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %esi
; X86-BMI1-NEXT:    movl 4(%eax), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB44_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB44_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %ebp
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    shrl %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB44_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %ebp
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB44_4:
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    andl %ebp, %esi
; X86-BMI1-NEXT:    andl %ebx, %edi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_c3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl (%eax), %esi
; X86-BMI2-NEXT:    movl 4(%eax), %eax
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB44_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB44_2:
; X86-BMI2-NEXT:    movb $64, %al
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl $-1, %ebp
; X86-BMI2-NEXT:    shrxl %eax, %ebp, %ebx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB44_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ebx, %ebp
; X86-BMI2-NEXT:    xorl %ebx, %ebx
; X86-BMI2-NEXT:  .LBB44_4:
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    andl %ebp, %esi
; X86-BMI2-NEXT:    andl %ebx, %edi
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    movl %edi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_c3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %r14
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movq (%rdi), %r14
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %r14
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movq $-1, %rbx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    andq %r14, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %r14
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_c3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %r14
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movl %esi, %ecx
; X64-BMI1-NEXT:    movq (%rdi), %r14
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-BMI1-NEXT:    shrq %cl, %r14
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movq $-1, %rbx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrq %cl, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    andq %r14, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %r14
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_c3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %r14
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movl %edx, %ebx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %r14
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movq $-1, %rcx
; X64-BMI2-NEXT:    shrxq %rax, %rcx, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    bzhiq %rbx, %r14, %rax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %r14
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %numhighbits = sub i8 64, %numlowbits
  %sh_prom = zext i8 %numhighbits to i64
  %mask = lshr i64 -1, %sh_prom
  call void @use64(i64 %mask)
  %masked = and i64 %mask, %shifted
  ret i64 %masked
}

define i64 @bextr64_c4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_c4_commutative:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB45_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB45_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %ebp
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    shrl %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB45_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %ebx, %ebp
; X86-NOBMI-NEXT:    xorl %ebx, %ebx
; X86-NOBMI-NEXT:  .LBB45_4:
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    andl %ebp, %esi
; X86-NOBMI-NEXT:    andl %ebx, %edi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_c4_commutative:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB45_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB45_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %ebp
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    shrl %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB45_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebx, %ebp
; X86-BMI1-NEXT:    xorl %ebx, %ebx
; X86-BMI1-NEXT:  .LBB45_4:
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    andl %ebp, %esi
; X86-BMI1-NEXT:    andl %ebx, %edi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_c4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB45_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB45_2:
; X86-BMI2-NEXT:    movb $64, %al
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl $-1, %ebp
; X86-BMI2-NEXT:    shrxl %eax, %ebp, %ebx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB45_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ebx, %ebp
; X86-BMI2-NEXT:    xorl %ebx, %ebx
; X86-BMI2-NEXT:  .LBB45_4:
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    andl %ebp, %esi
; X86-BMI2-NEXT:    andl %ebx, %edi
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    movl %edi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_c4_commutative:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %r14
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    pushq %rax
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %r14
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %r14
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movq $-1, %rbx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    andq %r14, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    addq $8, %rsp
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %r14
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_c4_commutative:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %r14
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    pushq %rax
; X64-BMI1-NEXT:    movq %rsi, %rcx
; X64-BMI1-NEXT:    movq %rdi, %r14
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-BMI1-NEXT:    shrq %cl, %r14
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movq $-1, %rbx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrq %cl, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    andq %r14, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    addq $8, %rsp
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %r14
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_c4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %r14
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    pushq %rax
; X64-BMI2-NEXT:    movq %rdx, %rbx
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %r14
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movq $-1, %rcx
; X64-BMI2-NEXT:    shrxq %rax, %rcx, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    bzhiq %rbx, %r14, %rax
; X64-BMI2-NEXT:    addq $8, %rsp
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %r14
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %mask = lshr i64 -1, %numhighbits
  call void @use64(i64 %mask)
  %masked = and i64 %shifted, %mask ; swapped order
  ret i64 %masked
}

define i64 @bextr64_c5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_c5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    shrl %cl, %edi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB46_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    xorl %edi, %edi
; X86-NOBMI-NEXT:  .LBB46_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %ebx
; X86-NOBMI-NEXT:    movl $-1, %ebp
; X86-NOBMI-NEXT:    shrl %cl, %ebp
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB46_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %ebp, %ebx
; X86-NOBMI-NEXT:    xorl %ebp, %ebp
; X86-NOBMI-NEXT:  .LBB46_4:
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    andl %ebx, %esi
; X86-NOBMI-NEXT:    andl %ebp, %edi
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-NOBMI-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_c5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    shrl %cl, %edi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB46_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    xorl %edi, %edi
; X86-BMI1-NEXT:  .LBB46_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %ebx
; X86-BMI1-NEXT:    movl $-1, %ebp
; X86-BMI1-NEXT:    shrl %cl, %ebp
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB46_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %ebp, %ebx
; X86-BMI1-NEXT:    xorl %ebp, %ebp
; X86-BMI1-NEXT:  .LBB46_4:
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    andl %ebx, %esi
; X86-BMI1-NEXT:    andl %ebp, %edi
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-BMI1-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_c5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $12, %esp
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB46_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    xorl %edi, %edi
; X86-BMI2-NEXT:  .LBB46_2:
; X86-BMI2-NEXT:    movb $64, %al
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl $-1, %ebp
; X86-BMI2-NEXT:    shrxl %eax, %ebp, %ebx
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB46_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ebx, %ebp
; X86-BMI2-NEXT:    xorl %ebx, %ebx
; X86-BMI2-NEXT:  .LBB46_4:
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %ebp
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    andl %ebp, %esi
; X86-BMI2-NEXT:    andl %ebx, %edi
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-BMI2-NEXT:    pushl {{[0-9]+}}(%esp)
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    movl %edi, %edx
; X86-BMI2-NEXT:    addl $12, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    popl %ebp
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_c5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %r15
; X64-NOBMI-NEXT:    pushq %r14
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movq %rsi, %r14
; X64-NOBMI-NEXT:    movq %rdi, %r15
; X64-NOBMI-NEXT:    movl %r14d, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %r15
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movq $-1, %rbx
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movq %rbx, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    andq %r15, %rbx
; X64-NOBMI-NEXT:    movq %r14, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    popq %r14
; X64-NOBMI-NEXT:    popq %r15
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_c5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %r15
; X64-BMI1-NEXT:    pushq %r14
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    movq %rsi, %r14
; X64-BMI1-NEXT:    movq %rdi, %r15
; X64-BMI1-NEXT:    movl %r14d, %ecx
; X64-BMI1-NEXT:    shrq %cl, %r15
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movq $-1, %rbx
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrq %cl, %rbx
; X64-BMI1-NEXT:    movq %rbx, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    andq %r15, %rbx
; X64-BMI1-NEXT:    movq %r14, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    popq %r14
; X64-BMI1-NEXT:    popq %r15
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_c5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %r15
; X64-BMI2-NEXT:    pushq %r14
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    movq %rdx, %rbx
; X64-BMI2-NEXT:    movq %rsi, %r14
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %r15
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    negb %al
; X64-BMI2-NEXT:    movq $-1, %rcx
; X64-BMI2-NEXT:    shrxq %rax, %rcx, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    bzhiq %rbx, %r15, %rbx
; X64-BMI2-NEXT:    movq %r14, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    movq %rbx, %rax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    popq %r14
; X64-BMI2-NEXT:    popq %r15
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %mask = lshr i64 -1, %numhighbits
  call void @use64(i64 %mask)
  %masked = and i64 %mask, %shifted
  call void @use64(i64 %numskipbits)
  ret i64 %masked
}

; 64-bit, but with 32-bit output

; Everything done in 64-bit, truncation happens last.
define i32 @bextr64_32_c0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_c0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %edx
; X86-NOBMI-NEXT:    shrl %cl, %edx
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB47_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:  .LBB47_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB47_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl $-1, %eax
; X86-NOBMI-NEXT:  .LBB47_4:
; X86-NOBMI-NEXT:    andl %edx, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_c0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %esi, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB47_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:  .LBB47_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl $-1, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB47_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl $-1, %eax
; X86-BMI1-NEXT:  .LBB47_4:
; X86-BMI1-NEXT:    andl %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_c0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB47_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
; X86-BMI2-NEXT:  .LBB47_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl $-1, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB47_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:  .LBB47_4:
; X86-BMI2-NEXT:    andl %edx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_c0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movq $-1, %rax
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_c0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_c0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %mask = lshr i64 -1, %numhighbits
  %masked = and i64 %mask, %shifted
  %res = trunc i64 %masked to i32
  ret i32 %res
}

; Shifting happens in 64-bit, then truncation. Masking is 32-bit.
define i32 @bextr64_32_c1(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_c1:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %edx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB48_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:  .LBB48_2:
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_c1:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB48_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB48_2:
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_c1:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB48_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB48_2:
; X86-BMI2-NEXT:    bzhil %eax, %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_c1:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_c1:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_c1:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %truncshifted = trunc i64 %shifted to i32
  %numhighbits = sub i32 32, %numlowbits
  %mask = lshr i32 -1, %numhighbits
  %masked = and i32 %mask, %truncshifted
  ret i32 %masked
}

; Shifting happens in 64-bit. Mask is 32-bit, but extended to 64-bit.
; Masking is 64-bit. Then truncation.
define i32 @bextr64_32_c2(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_c2:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %edx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB49_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:  .LBB49_2:
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_c2:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB49_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB49_2:
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_c2:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB49_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB49_2:
; X86-BMI2-NEXT:    bzhil %eax, %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_c2:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_c2:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_c2:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %mask = lshr i32 -1, %numhighbits
  %zextmask = zext i32 %mask to i64
  %masked = and i64 %zextmask, %shifted
  %truncmasked = trunc i64 %masked to i32
  ret i32 %truncmasked
}

; Shifting happens in 64-bit. Mask is 32-bit, but calculated in 64-bit.
; Masking is 64-bit. Then truncation.
define i32 @bextr64_32_c3(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_c3:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %edx
; X86-NOBMI-NEXT:    shrl %cl, %edx
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB50_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:  .LBB50_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:    movl $-1, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB50_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:  .LBB50_4:
; X86-NOBMI-NEXT:    andl %edx, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_c3:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %esi, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB50_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:  .LBB50_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:    movl $-1, %esi
; X86-BMI1-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB50_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:  .LBB50_4:
; X86-BMI1-NEXT:    andl %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_c3:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrdl %cl, %eax, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB50_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
; X86-BMI2-NEXT:  .LBB50_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:    movl $-1, %esi
; X86-BMI2-NEXT:    shrdl %cl, %eax, %esi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    jne .LBB50_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:  .LBB50_4:
; X86-BMI2-NEXT:    andl %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_c3:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rdi
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    andl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_c3:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    movq %rsi, %rcx
; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-BMI1-NEXT:    shrq %cl, %rdi
; X64-BMI1-NEXT:    negb %dl
; X64-BMI1-NEXT:    movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-BMI1-NEXT:    movl %edx, %ecx
; X64-BMI1-NEXT:    shrq %cl, %rax
; X64-BMI1-NEXT:    andl %edi, %eax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_c3:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rcx
; X64-BMI2-NEXT:    negb %dl
; X64-BMI2-NEXT:    movl $4294967295, %eax # imm = 0xFFFFFFFF
; X64-BMI2-NEXT:    shrxq %rdx, %rax, %rax
; X64-BMI2-NEXT:    andl %ecx, %eax
; X64-BMI2-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %mask = lshr i64 4294967295, %numhighbits
  %masked = and i64 %mask, %shifted
  %truncmasked = trunc i64 %masked to i32
  ret i32 %truncmasked
}

; ---------------------------------------------------------------------------- ;
; Pattern d. 32-bit.
; ---------------------------------------------------------------------------- ;

define i32 @bextr32_d0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_d0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_d0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_d0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_d0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_d0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_d0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %highbitscleared = shl i32 %shifted, %numhighbits
  %masked = lshr i32 %highbitscleared, %numhighbits
  ret i32 %masked
}

define i32 @bextr32_d1_indexzext(i32 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_d1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_d1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    orl %eax, %ecx
; X86-BMI1-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_d1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_d1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl %edi, %eax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_d1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_d1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %numhighbits = sub i8 32, %numlowbits
  %sh_prom = zext i8 %numhighbits to i32
  %highbitscleared = shl i32 %shifted, %sh_prom
  %masked = lshr i32 %highbitscleared, %sh_prom
  ret i32 %masked
}

define i32 @bextr32_d2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_d2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_d2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, (%eax), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_d2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_d2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %eax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_d2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, (%rdi), %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_d2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %shifted = lshr i32 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %highbitscleared = shl i32 %shifted, %numhighbits
  %masked = lshr i32 %highbitscleared, %numhighbits
  ret i32 %masked
}

define i32 @bextr32_d3_load_indexzext(i32* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_d3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_d3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, (%eax), %eax
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_d3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
; X86-BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
; X86-BMI2-NEXT:    bzhil %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_d3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movl (%rdi), %eax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_d3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, (%rdi), %eax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_d3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, i32* %w
  %skip = zext i8 %numskipbits to i32
  %shifted = lshr i32 %val, %skip
  %numhighbits = sub i8 32, %numlowbits
  %sh_prom = zext i8 %numhighbits to i32
  %highbitscleared = shl i32 %shifted, %sh_prom
  %masked = lshr i32 %highbitscleared, %sh_prom
  ret i32 %masked
}

define i32 @bextr32_d5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr32_d5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %eax, %ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %esi
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    movl %eax, (%esp)
; X86-NOBMI-NEXT:    calll use32@PLT
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    addl $8, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr32_d5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    shll $8, %ecx
; X86-BMI1-NEXT:    movzbl %al, %edx
; X86-BMI1-NEXT:    orl %ecx, %edx
; X86-BMI1-NEXT:    bextrl %edx, {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl %eax, (%esp)
; X86-BMI1-NEXT:    calll use32@PLT
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    addl $8, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr32_d5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    bzhil %eax, %edx, %esi
; X86-BMI2-NEXT:    movl %ecx, (%esp)
; X86-BMI2-NEXT:    calll use32@PLT
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    addl $8, %esp
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr32_d5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movl %edi, %ebx
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %ebx
; X64-NOBMI-NEXT:    shrl %cl, %ebx
; X64-NOBMI-NEXT:    movl %esi, %edi
; X64-NOBMI-NEXT:    callq use32@PLT
; X64-NOBMI-NEXT:    movl %ebx, %eax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr32_d5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrl %eax, %edi, %ebx
; X64-BMI1-NEXT:    movl %esi, %edi
; X64-BMI1-NEXT:    callq use32@PLT
; X64-BMI1-NEXT:    movl %ebx, %eax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr32_d5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %ebx
; X64-BMI2-NEXT:    movl %esi, %edi
; X64-BMI2-NEXT:    callq use32@PLT
; X64-BMI2-NEXT:    movl %ebx, %eax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    retq
  %shifted = lshr i32 %val, %numskipbits
  %numhighbits = sub i32 32, %numlowbits
  %highbitscleared = shl i32 %shifted, %numhighbits
  %masked = lshr i32 %highbitscleared, %numhighbits
  call void @use32(i32 %numskipbits)
  ret i32 %masked
}

; 64-bit.

define i64 @bextr64_d0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_d0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB56_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB56_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
; X86-NOBMI-NEXT:    shll %cl, %edi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl %edi, %ebx
; X86-NOBMI-NEXT:    jne .LBB56_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %ebx
; X86-NOBMI-NEXT:  .LBB56_4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl $0, %edx
; X86-NOBMI-NEXT:    jne .LBB56_6
; X86-NOBMI-NEXT:  # %bb.5:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:  .LBB56_6:
; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB56_8
; X86-NOBMI-NEXT:  # %bb.7:
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:  .LBB56_8:
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_d0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    movl %edx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    shrdl %cl, %edx, %edi
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB56_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB56_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shldl %cl, %edi, %eax
; X86-BMI1-NEXT:    shll %cl, %edi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl %edi, %ebx
; X86-BMI1-NEXT:    jne .LBB56_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %ebx
; X86-BMI1-NEXT:  .LBB56_4:
; X86-BMI1-NEXT:    movl %ebx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl $0, %edx
; X86-BMI1-NEXT:    jne .LBB56_6
; X86-BMI1-NEXT:  # %bb.5:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:  .LBB56_6:
; X86-BMI1-NEXT:    shrdl %cl, %ebx, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB56_8
; X86-BMI1-NEXT:  # %bb.7:
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:  .LBB56_8:
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_d0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %esi
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB56_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    xorl %esi, %esi
; X86-BMI2-NEXT:  .LBB56_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shldl %cl, %eax, %esi
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB56_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    movl $0, %edi
; X86-BMI2-NEXT:  .LBB56_4:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %eax
; X86-BMI2-NEXT:    jne .LBB56_6
; X86-BMI2-NEXT:  # %bb.5:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:  .LBB56_6:
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    jne .LBB56_8
; X86-BMI2-NEXT:  # %bb.7:
; X86-BMI2-NEXT:    movl %edi, %eax
; X86-BMI2-NEXT:  .LBB56_8:
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_d0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_d0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_d0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %highbitscleared = shl i64 %shifted, %numhighbits
  %masked = lshr i64 %highbitscleared, %numhighbits
  ret i64 %masked
}

define i64 @bextr64_d1_indexzext(i64 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_d1_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB57_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB57_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
; X86-NOBMI-NEXT:    shll %cl, %edi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl %edi, %ebx
; X86-NOBMI-NEXT:    jne .LBB57_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %ebx
; X86-NOBMI-NEXT:  .LBB57_4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl $0, %edx
; X86-NOBMI-NEXT:    jne .LBB57_6
; X86-NOBMI-NEXT:  # %bb.5:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:  .LBB57_6:
; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB57_8
; X86-NOBMI-NEXT:  # %bb.7:
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:  .LBB57_8:
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_d1_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    movl %edx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    shrdl %cl, %edx, %edi
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB57_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB57_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shldl %cl, %edi, %eax
; X86-BMI1-NEXT:    shll %cl, %edi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl %edi, %ebx
; X86-BMI1-NEXT:    jne .LBB57_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %ebx
; X86-BMI1-NEXT:  .LBB57_4:
; X86-BMI1-NEXT:    movl %ebx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl $0, %edx
; X86-BMI1-NEXT:    jne .LBB57_6
; X86-BMI1-NEXT:  # %bb.5:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:  .LBB57_6:
; X86-BMI1-NEXT:    shrdl %cl, %ebx, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB57_8
; X86-BMI1-NEXT:  # %bb.7:
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:  .LBB57_8:
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_d1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %esi
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB57_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    xorl %esi, %esi
; X86-BMI2-NEXT:  .LBB57_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shldl %cl, %eax, %esi
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB57_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    movl $0, %edi
; X86-BMI2-NEXT:  .LBB57_4:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %eax
; X86-BMI2-NEXT:    jne .LBB57_6
; X86-BMI2-NEXT:  # %bb.5:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:  .LBB57_6:
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    jne .LBB57_8
; X86-BMI2-NEXT:  # %bb.7:
; X86-BMI2-NEXT:    movl %edi, %eax
; X86-BMI2-NEXT:  .LBB57_8:
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_d1_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movq %rdi, %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_d1_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_d1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %numhighbits = sub i8 64, %numlowbits
  %sh_prom = zext i8 %numhighbits to i64
  %highbitscleared = shl i64 %shifted, %sh_prom
  %masked = lshr i64 %highbitscleared, %sh_prom
  ret i64 %masked
}

define i64 @bextr64_d2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_d2_load:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %edi
; X86-NOBMI-NEXT:    movl 4(%eax), %edx
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB58_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB58_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
; X86-NOBMI-NEXT:    shll %cl, %edi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl %edi, %ebx
; X86-NOBMI-NEXT:    jne .LBB58_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %ebx
; X86-NOBMI-NEXT:  .LBB58_4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl $0, %edx
; X86-NOBMI-NEXT:    jne .LBB58_6
; X86-NOBMI-NEXT:  # %bb.5:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:  .LBB58_6:
; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB58_8
; X86-NOBMI-NEXT:  # %bb.7:
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:  .LBB58_8:
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_d2_load:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %edi
; X86-BMI1-NEXT:    movl 4(%eax), %edx
; X86-BMI1-NEXT:    movl %edx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    shrdl %cl, %edx, %edi
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB58_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB58_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shldl %cl, %edi, %eax
; X86-BMI1-NEXT:    shll %cl, %edi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl %edi, %ebx
; X86-BMI1-NEXT:    jne .LBB58_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %ebx
; X86-BMI1-NEXT:  .LBB58_4:
; X86-BMI1-NEXT:    movl %ebx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl $0, %edx
; X86-BMI1-NEXT:    jne .LBB58_6
; X86-BMI1-NEXT:  # %bb.5:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:  .LBB58_6:
; X86-BMI1-NEXT:    shrdl %cl, %ebx, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB58_8
; X86-BMI1-NEXT:  # %bb.7:
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:  .LBB58_8:
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_d2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl (%edx), %eax
; X86-BMI2-NEXT:    movl 4(%edx), %edx
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %esi
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB58_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    xorl %esi, %esi
; X86-BMI2-NEXT:  .LBB58_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shldl %cl, %eax, %esi
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB58_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    movl $0, %edi
; X86-BMI2-NEXT:  .LBB58_4:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %eax
; X86-BMI2-NEXT:    jne .LBB58_6
; X86-BMI2-NEXT:  # %bb.5:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:  .LBB58_6:
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    jne .LBB58_8
; X86-BMI2-NEXT:  # %bb.7:
; X86-BMI2-NEXT:    movl %edi, %eax
; X86-BMI2-NEXT:  .LBB58_8:
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_d2_load:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq (%rdi), %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_d2_load:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, (%rdi), %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_d2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %highbitscleared = shl i64 %shifted, %numhighbits
  %masked = lshr i64 %highbitscleared, %numhighbits
  ret i64 %masked
}

define i64 @bextr64_d3_load_indexzext(i64* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_d3_load_indexzext:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl (%eax), %edi
; X86-NOBMI-NEXT:    movl 4(%eax), %edx
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB59_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %edi
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB59_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
; X86-NOBMI-NEXT:    shll %cl, %edi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl %edi, %ebx
; X86-NOBMI-NEXT:    jne .LBB59_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %eax, %ebx
; X86-NOBMI-NEXT:  .LBB59_4:
; X86-NOBMI-NEXT:    movl %ebx, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl $0, %edx
; X86-NOBMI-NEXT:    jne .LBB59_6
; X86-NOBMI-NEXT:  # %bb.5:
; X86-NOBMI-NEXT:    movl %edi, %esi
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:  .LBB59_6:
; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB59_8
; X86-NOBMI-NEXT:  # %bb.7:
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:  .LBB59_8:
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_d3_load_indexzext:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl (%eax), %edi
; X86-BMI1-NEXT:    movl 4(%eax), %edx
; X86-BMI1-NEXT:    movl %edx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    shrdl %cl, %edx, %edi
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB59_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %edi
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB59_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shldl %cl, %edi, %eax
; X86-BMI1-NEXT:    shll %cl, %edi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl %edi, %ebx
; X86-BMI1-NEXT:    jne .LBB59_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %eax, %ebx
; X86-BMI1-NEXT:  .LBB59_4:
; X86-BMI1-NEXT:    movl %ebx, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl $0, %edx
; X86-BMI1-NEXT:    jne .LBB59_6
; X86-BMI1-NEXT:  # %bb.5:
; X86-BMI1-NEXT:    movl %edi, %esi
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:  .LBB59_6:
; X86-BMI1-NEXT:    shrdl %cl, %ebx, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB59_8
; X86-BMI1-NEXT:  # %bb.7:
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:  .LBB59_8:
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_d3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl (%edx), %eax
; X86-BMI2-NEXT:    movl 4(%edx), %edx
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %esi
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB59_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    xorl %esi, %esi
; X86-BMI2-NEXT:  .LBB59_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shldl %cl, %eax, %esi
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB59_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:    movl $0, %edi
; X86-BMI2-NEXT:  .LBB59_4:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %eax
; X86-BMI2-NEXT:    jne .LBB59_6
; X86-BMI2-NEXT:  # %bb.5:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:  .LBB59_6:
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edi
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    jne .LBB59_8
; X86-BMI2-NEXT:  # %bb.7:
; X86-BMI2-NEXT:    movl %edi, %eax
; X86-BMI2-NEXT:  .LBB59_8:
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_d3_load_indexzext:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    movq (%rdi), %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_d3_load_indexzext:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, (%rdi), %rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_d3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, i64* %w
  %skip = zext i8 %numskipbits to i64
  %shifted = lshr i64 %val, %skip
  %numhighbits = sub i8 64, %numlowbits
  %sh_prom = zext i8 %numhighbits to i64
  %highbitscleared = shl i64 %shifted, %sh_prom
  %masked = lshr i64 %highbitscleared, %sh_prom
  ret i64 %masked
}

define i64 @bextr64_d5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_d5_skipextrauses:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %ebp
; X86-NOBMI-NEXT:    pushl %ebx
; X86-NOBMI-NEXT:    pushl %edi
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    subl $12, %esp
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl %edx, %esi
; X86-NOBMI-NEXT:    movl %eax, %ecx
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    shrdl %cl, %edx, %ebx
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:    testb $32, %al
; X86-NOBMI-NEXT:    je .LBB60_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %esi, %ebx
; X86-NOBMI-NEXT:    xorl %esi, %esi
; X86-NOBMI-NEXT:  .LBB60_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shldl %cl, %ebx, %esi
; X86-NOBMI-NEXT:    shll %cl, %ebx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl %ebx, %ebp
; X86-NOBMI-NEXT:    jne .LBB60_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %esi, %ebp
; X86-NOBMI-NEXT:  .LBB60_4:
; X86-NOBMI-NEXT:    movl %ebp, %esi
; X86-NOBMI-NEXT:    shrl %cl, %esi
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl $0, %edi
; X86-NOBMI-NEXT:    jne .LBB60_6
; X86-NOBMI-NEXT:  # %bb.5:
; X86-NOBMI-NEXT:    movl %ebx, %edx
; X86-NOBMI-NEXT:    movl %esi, %edi
; X86-NOBMI-NEXT:  .LBB60_6:
; X86-NOBMI-NEXT:    shrdl %cl, %ebp, %edx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI-NEXT:    jne .LBB60_8
; X86-NOBMI-NEXT:  # %bb.7:
; X86-NOBMI-NEXT:    movl %edx, %esi
; X86-NOBMI-NEXT:  .LBB60_8:
; X86-NOBMI-NEXT:    subl $8, %esp
; X86-NOBMI-NEXT:    pushl %ecx
; X86-NOBMI-NEXT:    pushl %eax
; X86-NOBMI-NEXT:    calll use64@PLT
; X86-NOBMI-NEXT:    addl $16, %esp
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    movl %edi, %edx
; X86-NOBMI-NEXT:    addl $12, %esp
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    popl %edi
; X86-NOBMI-NEXT:    popl %ebx
; X86-NOBMI-NEXT:    popl %ebp
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_d5_skipextrauses:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %ebp
; X86-BMI1-NEXT:    pushl %ebx
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    subl $12, %esp
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %ebx
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:    movl %eax, %ecx
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    shrdl %cl, %edx, %ebx
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:    testb $32, %al
; X86-BMI1-NEXT:    je .LBB60_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %ebx
; X86-BMI1-NEXT:    xorl %esi, %esi
; X86-BMI1-NEXT:  .LBB60_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shldl %cl, %ebx, %esi
; X86-BMI1-NEXT:    shll %cl, %ebx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl %ebx, %ebp
; X86-BMI1-NEXT:    jne .LBB60_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %esi, %ebp
; X86-BMI1-NEXT:  .LBB60_4:
; X86-BMI1-NEXT:    movl %ebp, %esi
; X86-BMI1-NEXT:    shrl %cl, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl $0, %edi
; X86-BMI1-NEXT:    jne .LBB60_6
; X86-BMI1-NEXT:  # %bb.5:
; X86-BMI1-NEXT:    movl %ebx, %edx
; X86-BMI1-NEXT:    movl %esi, %edi
; X86-BMI1-NEXT:  .LBB60_6:
; X86-BMI1-NEXT:    shrdl %cl, %ebp, %edx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI1-NEXT:    jne .LBB60_8
; X86-BMI1-NEXT:  # %bb.7:
; X86-BMI1-NEXT:    movl %edx, %esi
; X86-BMI1-NEXT:  .LBB60_8:
; X86-BMI1-NEXT:    subl $8, %esp
; X86-BMI1-NEXT:    pushl %ecx
; X86-BMI1-NEXT:    pushl %eax
; X86-BMI1-NEXT:    calll use64@PLT
; X86-BMI1-NEXT:    addl $16, %esp
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    addl $12, %esp
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    popl %ebx
; X86-BMI1-NEXT:    popl %ebp
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_d5_skipextrauses:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %edi
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl %eax, %ecx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %edi
; X86-BMI2-NEXT:    shrxl %eax, %edx, %edx
; X86-BMI2-NEXT:    xorl %esi, %esi
; X86-BMI2-NEXT:    testb $32, %al
; X86-BMI2-NEXT:    je .LBB60_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %edi
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB60_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shldl %cl, %edi, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %edi, %ebx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB60_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %ebx, %edx
; X86-BMI2-NEXT:    movl $0, %ebx
; X86-BMI2-NEXT:  .LBB60_4:
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edi
; X86-BMI2-NEXT:    jne .LBB60_6
; X86-BMI2-NEXT:  # %bb.5:
; X86-BMI2-NEXT:    movl %edi, %esi
; X86-BMI2-NEXT:  .LBB60_6:
; X86-BMI2-NEXT:    shrdl %cl, %edx, %ebx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    jne .LBB60_8
; X86-BMI2-NEXT:  # %bb.7:
; X86-BMI2-NEXT:    movl %ebx, %edi
; X86-BMI2-NEXT:  .LBB60_8:
; X86-BMI2-NEXT:    subl $8, %esp
; X86-BMI2-NEXT:    pushl %ecx
; X86-BMI2-NEXT:    pushl %eax
; X86-BMI2-NEXT:    calll use64@PLT
; X86-BMI2-NEXT:    addl $16, %esp
; X86-BMI2-NEXT:    movl %edi, %eax
; X86-BMI2-NEXT:    movl %esi, %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %edi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_d5_skipextrauses:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    pushq %rbx
; X64-NOBMI-NEXT:    movq %rdi, %rbx
; X64-NOBMI-NEXT:    movl %esi, %ecx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rbx
; X64-NOBMI-NEXT:    shrq %cl, %rbx
; X64-NOBMI-NEXT:    movq %rsi, %rdi
; X64-NOBMI-NEXT:    callq use64@PLT
; X64-NOBMI-NEXT:    movq %rbx, %rax
; X64-NOBMI-NEXT:    popq %rbx
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_d5_skipextrauses:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    pushq %rbx
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rbx
; X64-BMI1-NEXT:    movq %rsi, %rdi
; X64-BMI1-NEXT:    callq use64@PLT
; X64-BMI1-NEXT:    movq %rbx, %rax
; X64-BMI1-NEXT:    popq %rbx
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_d5_skipextrauses:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    pushq %rbx
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rbx
; X64-BMI2-NEXT:    movq %rsi, %rdi
; X64-BMI2-NEXT:    callq use64@PLT
; X64-BMI2-NEXT:    movq %rbx, %rax
; X64-BMI2-NEXT:    popq %rbx
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %highbitscleared = shl i64 %shifted, %numhighbits
  %masked = lshr i64 %highbitscleared, %numhighbits
  call void @use64(i64 %numskipbits)
  ret i64 %masked
}

; 64-bit, but with 32-bit output

; Everything done in 64-bit, truncation happens last.
define i32 @bextr64_32_d0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_d0:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %edx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB61_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %eax, %edx
; X86-NOBMI-NEXT:    xorl %eax, %eax
; X86-NOBMI-NEXT:  .LBB61_2:
; X86-NOBMI-NEXT:    movb $64, %cl
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shldl %cl, %edx, %eax
; X86-NOBMI-NEXT:    shll %cl, %edx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    je .LBB61_4
; X86-NOBMI-NEXT:  # %bb.3:
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:  .LBB61_4:
; X86-NOBMI-NEXT:    shrdl %cl, %eax, %edx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB61_6
; X86-NOBMI-NEXT:  # %bb.5:
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:  .LBB61_6:
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_d0:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl %esi, %eax
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB61_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %eax, %edx
; X86-BMI1-NEXT:    xorl %eax, %eax
; X86-BMI1-NEXT:  .LBB61_2:
; X86-BMI1-NEXT:    movb $64, %cl
; X86-BMI1-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    shldl %cl, %edx, %eax
; X86-BMI1-NEXT:    shll %cl, %edx
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    je .LBB61_4
; X86-BMI1-NEXT:  # %bb.3:
; X86-BMI1-NEXT:    movl %edx, %eax
; X86-BMI1-NEXT:    xorl %edx, %edx
; X86-BMI1-NEXT:  .LBB61_4:
; X86-BMI1-NEXT:    shrdl %cl, %eax, %edx
; X86-BMI1-NEXT:    shrl %cl, %eax
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB61_6
; X86-BMI1-NEXT:  # %bb.5:
; X86-BMI1-NEXT:    movl %edx, %eax
; X86-BMI1-NEXT:  .LBB61_6:
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_d0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB61_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %edx, %eax
; X86-BMI2-NEXT:    xorl %edx, %edx
; X86-BMI2-NEXT:  .LBB61_2:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shldl %cl, %eax, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB61_4
; X86-BMI2-NEXT:  # %bb.3:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB61_4:
; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB61_6
; X86-BMI2-NEXT:  # %bb.5:
; X86-BMI2-NEXT:    shrxl %ecx, %edx, %eax
; X86-BMI2-NEXT:  .LBB61_6:
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_d0:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shlq %cl, %rax
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_d0:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_d0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhiq %rdx, %rax, %rax
; X64-BMI2-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %numhighbits = sub i64 64, %numlowbits
  %highbitscleared = shl i64 %shifted, %numhighbits
  %masked = lshr i64 %highbitscleared, %numhighbits
  %res = trunc i64 %masked to i32
  ret i32 %res
}

; Shifting happens in 64-bit, then truncation. Masking is 32-bit.
define i32 @bextr64_32_d1(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bextr64_32_d1:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    pushl %esi
; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI-NEXT:    movl %esi, %eax
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    shrdl %cl, %esi, %edx
; X86-NOBMI-NEXT:    testb $32, %cl
; X86-NOBMI-NEXT:    jne .LBB62_2
; X86-NOBMI-NEXT:  # %bb.1:
; X86-NOBMI-NEXT:    movl %edx, %eax
; X86-NOBMI-NEXT:  .LBB62_2:
; X86-NOBMI-NEXT:    xorl %ecx, %ecx
; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI-NEXT:    shll %cl, %eax
; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT:    shrl %cl, %eax
; X86-NOBMI-NEXT:    popl %esi
; X86-NOBMI-NEXT:    retl
;
; X86-BMI1-LABEL: bextr64_32_d1:
; X86-BMI1:       # %bb.0:
; X86-BMI1-NEXT:    pushl %edi
; X86-BMI1-NEXT:    pushl %esi
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edi
; X86-BMI1-NEXT:    movl %edi, %edx
; X86-BMI1-NEXT:    shrl %cl, %edx
; X86-BMI1-NEXT:    shrdl %cl, %edi, %esi
; X86-BMI1-NEXT:    testb $32, %cl
; X86-BMI1-NEXT:    jne .LBB62_2
; X86-BMI1-NEXT:  # %bb.1:
; X86-BMI1-NEXT:    movl %esi, %edx
; X86-BMI1-NEXT:  .LBB62_2:
; X86-BMI1-NEXT:    shll $8, %eax
; X86-BMI1-NEXT:    bextrl %eax, %edx, %eax
; X86-BMI1-NEXT:    popl %esi
; X86-BMI1-NEXT:    popl %edi
; X86-BMI1-NEXT:    retl
;
; X86-BMI2-LABEL: bextr64_32_d1:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-BMI2-NEXT:    shrdl %cl, %esi, %edx
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB62_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edx
; X86-BMI2-NEXT:  .LBB62_2:
; X86-BMI2-NEXT:    bzhil %eax, %edx, %eax
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI-LABEL: bextr64_32_d1:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rsi, %rcx
; X64-NOBMI-NEXT:    movq %rdi, %rax
; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI-NEXT:    shrq %cl, %rax
; X64-NOBMI-NEXT:    negb %dl
; X64-NOBMI-NEXT:    movl %edx, %ecx
; X64-NOBMI-NEXT:    shll %cl, %eax
; X64-NOBMI-NEXT:    shrl %cl, %eax
; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-NOBMI-NEXT:    retq
;
; X64-BMI1-LABEL: bextr64_32_d1:
; X64-BMI1:       # %bb.0:
; X64-BMI1-NEXT:    shll $8, %edx
; X64-BMI1-NEXT:    movzbl %sil, %eax
; X64-BMI1-NEXT:    orl %edx, %eax
; X64-BMI1-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMI1-NEXT:    # kill: def $eax killed $eax killed $rax
; X64-BMI1-NEXT:    retq
;
; X64-BMI2-LABEL: bextr64_32_d1:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    bzhil %edx, %eax, %eax
; X64-BMI2-NEXT:    retq
  %shifted = lshr i64 %val, %numskipbits
  %truncshifted = trunc i64 %shifted to i32
  %numhighbits = sub i32 32, %numlowbits
  %highbitscleared = shl i32 %truncshifted, %numhighbits
  %masked = lshr i32 %highbitscleared, %numhighbits
  ret i32 %masked
}

; ---------------------------------------------------------------------------- ;
; Constant
; ---------------------------------------------------------------------------- ;

; https://bugs.llvm.org/show_bug.cgi?id=38938
define void @pr38938(i32* %a0, i64* %a1) nounwind {
; X86-NOBMI-LABEL: pr38938:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI-NEXT:    movl (%ecx), %ecx
; X86-NOBMI-NEXT:    shrl $19, %ecx
; X86-NOBMI-NEXT:    andl $4092, %ecx # imm = 0xFFC
; X86-NOBMI-NEXT:    incl (%eax,%ecx)
; X86-NOBMI-NEXT:    retl
;
; X86-BMINOTBM-LABEL: pr38938:
; X86-BMINOTBM:       # %bb.0:
; X86-BMINOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMINOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMINOTBM-NEXT:    movl $2581, %edx # imm = 0xA15
; X86-BMINOTBM-NEXT:    bextrl %edx, (%ecx), %ecx
; X86-BMINOTBM-NEXT:    incl (%eax,%ecx,4)
; X86-BMINOTBM-NEXT:    retl
;
; X86-BMITBM-LABEL: pr38938:
; X86-BMITBM:       # %bb.0:
; X86-BMITBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMITBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMITBM-NEXT:    bextrl $2581, (%ecx), %ecx # imm = 0xA15
; X86-BMITBM-NEXT:    incl (%eax,%ecx,4)
; X86-BMITBM-NEXT:    retl
;
; X64-NOBMI-LABEL: pr38938:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq (%rsi), %rax
; X64-NOBMI-NEXT:    shrq $19, %rax
; X64-NOBMI-NEXT:    andl $4092, %eax # imm = 0xFFC
; X64-NOBMI-NEXT:    incl (%rdi,%rax)
; X64-NOBMI-NEXT:    retq
;
; X64-BMINOTBM-LABEL: pr38938:
; X64-BMINOTBM:       # %bb.0:
; X64-BMINOTBM-NEXT:    movl $2581, %eax # imm = 0xA15
; X64-BMINOTBM-NEXT:    bextrq %rax, (%rsi), %rax
; X64-BMINOTBM-NEXT:    incl (%rdi,%rax,4)
; X64-BMINOTBM-NEXT:    retq
;
; X64-BMITBM-LABEL: pr38938:
; X64-BMITBM:       # %bb.0:
; X64-BMITBM-NEXT:    bextrq $2581, (%rsi), %rax # imm = 0xA15
; X64-BMITBM-NEXT:    incl (%rdi,%rax,4)
; X64-BMITBM-NEXT:    retq
  %tmp = load i64, i64* %a1, align 8
  %tmp1 = lshr i64 %tmp, 21
  %tmp2 = and i64 %tmp1, 1023
  %tmp3 = getelementptr inbounds i32, i32* %a0, i64 %tmp2
  %tmp4 = load i32, i32* %tmp3, align 4
  %tmp5 = add nsw i32 %tmp4, 1
  store i32 %tmp5, i32* %tmp3, align 4
  ret void
}

; The most canonical variant
define i32 @c0_i32(i32 %arg) nounwind {
; X86-NOBMI-LABEL: c0_i32:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    shrl $19, %eax
; X86-NOBMI-NEXT:    andl $1023, %eax # imm = 0x3FF
; X86-NOBMI-NEXT:    retl
;
; X86-BMINOTBM-LABEL: c0_i32:
; X86-BMINOTBM:       # %bb.0:
; X86-BMINOTBM-NEXT:    movl $2579, %eax # imm = 0xA13
; X86-BMINOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
; X86-BMINOTBM-NEXT:    retl
;
; X86-BMITBM-LABEL: c0_i32:
; X86-BMITBM:       # %bb.0:
; X86-BMITBM-NEXT:    bextrl $2579, {{[0-9]+}}(%esp), %eax # imm = 0xA13
; X86-BMITBM-NEXT:    retl
;
; X64-NOBMI-LABEL: c0_i32:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movl %edi, %eax
; X64-NOBMI-NEXT:    shrl $19, %eax
; X64-NOBMI-NEXT:    andl $1023, %eax # imm = 0x3FF
; X64-NOBMI-NEXT:    retq
;
; X64-BMINOTBM-LABEL: c0_i32:
; X64-BMINOTBM:       # %bb.0:
; X64-BMINOTBM-NEXT:    movl $2579, %eax # imm = 0xA13
; X64-BMINOTBM-NEXT:    bextrl %eax, %edi, %eax
; X64-BMINOTBM-NEXT:    retq
;
; X64-BMITBM-LABEL: c0_i32:
; X64-BMITBM:       # %bb.0:
; X64-BMITBM-NEXT:    bextrl $2579, %edi, %eax # imm = 0xA13
; X64-BMITBM-NEXT:    retq
  %tmp0 = lshr i32 %arg, 19
  %tmp1 = and i32 %tmp0, 1023
  ret i32 %tmp1
}

; Should be still fine, but the mask is shifted
define i32 @c1_i32(i32 %arg) nounwind {
; X86-LABEL: c1_i32:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl $19, %eax
; X86-NEXT:    andl $4092, %eax # imm = 0xFFC
; X86-NEXT:    retl
;
; X64-LABEL: c1_i32:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    shrl $19, %eax
; X64-NEXT:    andl $4092, %eax # imm = 0xFFC
; X64-NEXT:    retq
  %tmp0 = lshr i32 %arg, 19
  %tmp1 = and i32 %tmp0, 4092
  ret i32 %tmp1
}

; Should be still fine, but the result is shifted left afterwards
define i32 @c2_i32(i32 %arg) nounwind {
; X86-LABEL: c2_i32:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl $17, %eax
; X86-NEXT:    andl $4092, %eax # imm = 0xFFC
; X86-NEXT:    retl
;
; X64-LABEL: c2_i32:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    shrl $17, %eax
; X64-NEXT:    andl $4092, %eax # imm = 0xFFC
; X64-NEXT:    retq
  %tmp0 = lshr i32 %arg, 19
  %tmp1 = and i32 %tmp0, 1023
  %tmp2 = shl i32 %tmp1, 2
  ret i32 %tmp2
}

; The mask covers newly shifted-in bit
define i32 @c4_i32_bad(i32 %arg) nounwind {
; X86-LABEL: c4_i32_bad:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl $19, %eax
; X86-NEXT:    andl $-2, %eax
; X86-NEXT:    retl
;
; X64-LABEL: c4_i32_bad:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    shrl $19, %eax
; X64-NEXT:    andl $-2, %eax
; X64-NEXT:    retq
  %tmp0 = lshr i32 %arg, 19
  %tmp1 = and i32 %tmp0, 16382
  ret i32 %tmp1
}

; i64

; The most canonical variant
define i64 @c0_i64(i64 %arg) nounwind {
; X86-NOBMI-LABEL: c0_i64:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    shrl $19, %eax
; X86-NOBMI-NEXT:    andl $1023, %eax # imm = 0x3FF
; X86-NOBMI-NEXT:    xorl %edx, %edx
; X86-NOBMI-NEXT:    retl
;
; X86-BMINOTBM-LABEL: c0_i64:
; X86-BMINOTBM:       # %bb.0:
; X86-BMINOTBM-NEXT:    movl $2579, %eax # imm = 0xA13
; X86-BMINOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
; X86-BMINOTBM-NEXT:    xorl %edx, %edx
; X86-BMINOTBM-NEXT:    retl
;
; X86-BMITBM-LABEL: c0_i64:
; X86-BMITBM:       # %bb.0:
; X86-BMITBM-NEXT:    bextrl $2579, {{[0-9]+}}(%esp), %eax # imm = 0xA13
; X86-BMITBM-NEXT:    xorl %edx, %edx
; X86-BMITBM-NEXT:    retl
;
; X64-NOBMI-LABEL: c0_i64:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    movq %rdi, %rax
; X64-NOBMI-NEXT:    shrq $51, %rax
; X64-NOBMI-NEXT:    andl $1023, %eax # imm = 0x3FF
; X64-NOBMI-NEXT:    retq
;
; X64-BMINOTBM-LABEL: c0_i64:
; X64-BMINOTBM:       # %bb.0:
; X64-BMINOTBM-NEXT:    movl $2611, %eax # imm = 0xA33
; X64-BMINOTBM-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMINOTBM-NEXT:    retq
;
; X64-BMITBM-LABEL: c0_i64:
; X64-BMITBM:       # %bb.0:
; X64-BMITBM-NEXT:    bextrq $2611, %rdi, %rax # imm = 0xA33
; X64-BMITBM-NEXT:    retq
  %tmp0 = lshr i64 %arg, 51
  %tmp1 = and i64 %tmp0, 1023
  ret i64 %tmp1
}

; Should be still fine, but the mask is shifted
define i64 @c1_i64(i64 %arg) nounwind {
; X86-LABEL: c1_i64:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl $19, %eax
; X86-NEXT:    andl $4092, %eax # imm = 0xFFC
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    retl
;
; X64-LABEL: c1_i64:
; X64:       # %bb.0:
; X64-NEXT:    movq %rdi, %rax
; X64-NEXT:    shrq $51, %rax
; X64-NEXT:    andl $4092, %eax # imm = 0xFFC
; X64-NEXT:    retq
  %tmp0 = lshr i64 %arg, 51
  %tmp1 = and i64 %tmp0, 4092
  ret i64 %tmp1
}

; Should be still fine, but the result is shifted left afterwards
define i64 @c2_i64(i64 %arg) nounwind {
; X86-LABEL: c2_i64:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl $17, %eax
; X86-NEXT:    andl $4092, %eax # imm = 0xFFC
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    retl
;
; X64-LABEL: c2_i64:
; X64:       # %bb.0:
; X64-NEXT:    movq %rdi, %rax
; X64-NEXT:    shrq $49, %rax
; X64-NEXT:    andl $4092, %eax # imm = 0xFFC
; X64-NEXT:    retq
  %tmp0 = lshr i64 %arg, 51
  %tmp1 = and i64 %tmp0, 1023
  %tmp2 = shl i64 %tmp1, 2
  ret i64 %tmp2
}

; The mask covers newly shifted-in bit
define i64 @c4_i64_bad(i64 %arg) nounwind {
; X86-LABEL: c4_i64_bad:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrl $19, %eax
; X86-NEXT:    andl $-2, %eax
; X86-NEXT:    xorl %edx, %edx
; X86-NEXT:    retl
;
; X64-LABEL: c4_i64_bad:
; X64:       # %bb.0:
; X64-NEXT:    movq %rdi, %rax
; X64-NEXT:    shrq $51, %rax
; X64-NEXT:    andl $-2, %eax
; X64-NEXT:    retq
  %tmp0 = lshr i64 %arg, 51
  %tmp1 = and i64 %tmp0, 16382
  ret i64 %tmp1
}

; ---------------------------------------------------------------------------- ;
; Constant, storing the result afterwards.
; ---------------------------------------------------------------------------- ;

; i32

; The most canonical variant
define void @c5_i32(i32 %arg, i32* %ptr) nounwind {
; X86-NOBMI-LABEL: c5_i32:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI-NEXT:    shrl $19, %ecx
; X86-NOBMI-NEXT:    andl $1023, %ecx # imm = 0x3FF
; X86-NOBMI-NEXT:    movl %ecx, (%eax)
; X86-NOBMI-NEXT:    retl
;
; X86-BMINOTBM-LABEL: c5_i32:
; X86-BMINOTBM:       # %bb.0:
; X86-BMINOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMINOTBM-NEXT:    movl $2579, %ecx # imm = 0xA13
; X86-BMINOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMINOTBM-NEXT:    movl %ecx, (%eax)
; X86-BMINOTBM-NEXT:    retl
;
; X86-BMITBM-LABEL: c5_i32:
; X86-BMITBM:       # %bb.0:
; X86-BMITBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMITBM-NEXT:    bextrl $2579, {{[0-9]+}}(%esp), %ecx # imm = 0xA13
; X86-BMITBM-NEXT:    movl %ecx, (%eax)
; X86-BMITBM-NEXT:    retl
;
; X64-NOBMI-LABEL: c5_i32:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    shrl $19, %edi
; X64-NOBMI-NEXT:    andl $1023, %edi # imm = 0x3FF
; X64-NOBMI-NEXT:    movl %edi, (%rsi)
; X64-NOBMI-NEXT:    retq
;
; X64-BMINOTBM-LABEL: c5_i32:
; X64-BMINOTBM:       # %bb.0:
; X64-BMINOTBM-NEXT:    movl $2579, %eax # imm = 0xA13
; X64-BMINOTBM-NEXT:    bextrl %eax, %edi, %eax
; X64-BMINOTBM-NEXT:    movl %eax, (%rsi)
; X64-BMINOTBM-NEXT:    retq
;
; X64-BMITBM-LABEL: c5_i32:
; X64-BMITBM:       # %bb.0:
; X64-BMITBM-NEXT:    bextrl $2579, %edi, %eax # imm = 0xA13
; X64-BMITBM-NEXT:    movl %eax, (%rsi)
; X64-BMITBM-NEXT:    retq
  %tmp0 = lshr i32 %arg, 19
  %tmp1 = and i32 %tmp0, 1023
  store i32 %tmp1, i32* %ptr
  ret void
}

; Should be still fine, but the mask is shifted
define void @c6_i32(i32 %arg, i32* %ptr) nounwind {
; X86-NOBMI-LABEL: c6_i32:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI-NEXT:    shrl $19, %ecx
; X86-NOBMI-NEXT:    andl $4095, %ecx # imm = 0xFFF
; X86-NOBMI-NEXT:    movl %ecx, (%eax)
; X86-NOBMI-NEXT:    retl
;
; X86-BMINOTBM-LABEL: c6_i32:
; X86-BMINOTBM:       # %bb.0:
; X86-BMINOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMINOTBM-NEXT:    movl $3091, %ecx # imm = 0xC13
; X86-BMINOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMINOTBM-NEXT:    movl %ecx, (%eax)
; X86-BMINOTBM-NEXT:    retl
;
; X86-BMITBM-LABEL: c6_i32:
; X86-BMITBM:       # %bb.0:
; X86-BMITBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMITBM-NEXT:    bextrl $3091, {{[0-9]+}}(%esp), %ecx # imm = 0xC13
; X86-BMITBM-NEXT:    movl %ecx, (%eax)
; X86-BMITBM-NEXT:    retl
;
; X64-NOBMI-LABEL: c6_i32:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    shrl $19, %edi
; X64-NOBMI-NEXT:    andl $4095, %edi # imm = 0xFFF
; X64-NOBMI-NEXT:    movl %edi, (%rsi)
; X64-NOBMI-NEXT:    retq
;
; X64-BMINOTBM-LABEL: c6_i32:
; X64-BMINOTBM:       # %bb.0:
; X64-BMINOTBM-NEXT:    movl $3091, %eax # imm = 0xC13
; X64-BMINOTBM-NEXT:    bextrl %eax, %edi, %eax
; X64-BMINOTBM-NEXT:    movl %eax, (%rsi)
; X64-BMINOTBM-NEXT:    retq
;
; X64-BMITBM-LABEL: c6_i32:
; X64-BMITBM:       # %bb.0:
; X64-BMITBM-NEXT:    bextrl $3091, %edi, %eax # imm = 0xC13
; X64-BMITBM-NEXT:    movl %eax, (%rsi)
; X64-BMITBM-NEXT:    retq
  %tmp0 = lshr i32 %arg, 19
  %tmp1 = and i32 %tmp0, 4095
  store i32 %tmp1, i32* %ptr
  ret void
}

; Should be still fine, but the result is shifted left afterwards
define void @c7_i32(i32 %arg, i32* %ptr) nounwind {
; X86-LABEL: c7_i32:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    shrl $17, %ecx
; X86-NEXT:    andl $4092, %ecx # imm = 0xFFC
; X86-NEXT:    movl %ecx, (%eax)
; X86-NEXT:    retl
;
; X64-LABEL: c7_i32:
; X64:       # %bb.0:
; X64-NEXT:    shrl $17, %edi
; X64-NEXT:    andl $4092, %edi # imm = 0xFFC
; X64-NEXT:    movl %edi, (%rsi)
; X64-NEXT:    retq
  %tmp0 = lshr i32 %arg, 19
  %tmp1 = and i32 %tmp0, 1023
  %tmp2 = shl i32 %tmp1, 2
  store i32 %tmp2, i32* %ptr
  ret void
}

; i64

; The most canonical variant
define void @c5_i64(i64 %arg, i64* %ptr) nounwind {
; X86-NOBMI-LABEL: c5_i64:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI-NEXT:    shrl $19, %ecx
; X86-NOBMI-NEXT:    andl $1023, %ecx # imm = 0x3FF
; X86-NOBMI-NEXT:    movl %ecx, (%eax)
; X86-NOBMI-NEXT:    movl $0, 4(%eax)
; X86-NOBMI-NEXT:    retl
;
; X86-BMINOTBM-LABEL: c5_i64:
; X86-BMINOTBM:       # %bb.0:
; X86-BMINOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMINOTBM-NEXT:    movl $2579, %ecx # imm = 0xA13
; X86-BMINOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMINOTBM-NEXT:    movl %ecx, (%eax)
; X86-BMINOTBM-NEXT:    movl $0, 4(%eax)
; X86-BMINOTBM-NEXT:    retl
;
; X86-BMITBM-LABEL: c5_i64:
; X86-BMITBM:       # %bb.0:
; X86-BMITBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMITBM-NEXT:    bextrl $2579, {{[0-9]+}}(%esp), %ecx # imm = 0xA13
; X86-BMITBM-NEXT:    movl %ecx, (%eax)
; X86-BMITBM-NEXT:    movl $0, 4(%eax)
; X86-BMITBM-NEXT:    retl
;
; X64-NOBMI-LABEL: c5_i64:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    shrq $51, %rdi
; X64-NOBMI-NEXT:    andl $1023, %edi # imm = 0x3FF
; X64-NOBMI-NEXT:    movq %rdi, (%rsi)
; X64-NOBMI-NEXT:    retq
;
; X64-BMINOTBM-LABEL: c5_i64:
; X64-BMINOTBM:       # %bb.0:
; X64-BMINOTBM-NEXT:    movl $2611, %eax # imm = 0xA33
; X64-BMINOTBM-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMINOTBM-NEXT:    movq %rax, (%rsi)
; X64-BMINOTBM-NEXT:    retq
;
; X64-BMITBM-LABEL: c5_i64:
; X64-BMITBM:       # %bb.0:
; X64-BMITBM-NEXT:    bextrq $2611, %rdi, %rax # imm = 0xA33
; X64-BMITBM-NEXT:    movq %rax, (%rsi)
; X64-BMITBM-NEXT:    retq
  %tmp0 = lshr i64 %arg, 51
  %tmp1 = and i64 %tmp0, 1023
  store i64 %tmp1, i64* %ptr
  ret void
}

; Should be still fine, but the mask is shifted
define void @c6_i64(i64 %arg, i64* %ptr) nounwind {
; X86-NOBMI-LABEL: c6_i64:
; X86-NOBMI:       # %bb.0:
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI-NEXT:    shrl $19, %ecx
; X86-NOBMI-NEXT:    andl $4095, %ecx # imm = 0xFFF
; X86-NOBMI-NEXT:    movl %ecx, (%eax)
; X86-NOBMI-NEXT:    movl $0, 4(%eax)
; X86-NOBMI-NEXT:    retl
;
; X86-BMINOTBM-LABEL: c6_i64:
; X86-BMINOTBM:       # %bb.0:
; X86-BMINOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMINOTBM-NEXT:    movl $3091, %ecx # imm = 0xC13
; X86-BMINOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %ecx
; X86-BMINOTBM-NEXT:    movl %ecx, (%eax)
; X86-BMINOTBM-NEXT:    movl $0, 4(%eax)
; X86-BMINOTBM-NEXT:    retl
;
; X86-BMITBM-LABEL: c6_i64:
; X86-BMITBM:       # %bb.0:
; X86-BMITBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMITBM-NEXT:    bextrl $3091, {{[0-9]+}}(%esp), %ecx # imm = 0xC13
; X86-BMITBM-NEXT:    movl %ecx, (%eax)
; X86-BMITBM-NEXT:    movl $0, 4(%eax)
; X86-BMITBM-NEXT:    retl
;
; X64-NOBMI-LABEL: c6_i64:
; X64-NOBMI:       # %bb.0:
; X64-NOBMI-NEXT:    shrq $51, %rdi
; X64-NOBMI-NEXT:    andl $4095, %edi # imm = 0xFFF
; X64-NOBMI-NEXT:    movq %rdi, (%rsi)
; X64-NOBMI-NEXT:    retq
;
; X64-BMINOTBM-LABEL: c6_i64:
; X64-BMINOTBM:       # %bb.0:
; X64-BMINOTBM-NEXT:    movl $3123, %eax # imm = 0xC33
; X64-BMINOTBM-NEXT:    bextrq %rax, %rdi, %rax
; X64-BMINOTBM-NEXT:    movq %rax, (%rsi)
; X64-BMINOTBM-NEXT:    retq
;
; X64-BMITBM-LABEL: c6_i64:
; X64-BMITBM:       # %bb.0:
; X64-BMITBM-NEXT:    bextrq $3123, %rdi, %rax # imm = 0xC33
; X64-BMITBM-NEXT:    movq %rax, (%rsi)
; X64-BMITBM-NEXT:    retq
  %tmp0 = lshr i64 %arg, 51
  %tmp1 = and i64 %tmp0, 4095
  store i64 %tmp1, i64* %ptr
  ret void
}

; Should be still fine, but the result is shifted left afterwards
define void @c7_i64(i64 %arg, i64* %ptr) nounwind {
; X86-LABEL: c7_i64:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    shrl $17, %ecx
; X86-NEXT:    andl $4092, %ecx # imm = 0xFFC
; X86-NEXT:    movl %ecx, (%eax)
; X86-NEXT:    movl $0, 4(%eax)
; X86-NEXT:    retl
;
; X64-LABEL: c7_i64:
; X64:       # %bb.0:
; X64-NEXT:    shrq $49, %rdi
; X64-NEXT:    andl $4092, %edi # imm = 0xFFC
; X64-NEXT:    movq %rdi, (%rsi)
; X64-NEXT:    retq
  %tmp0 = lshr i64 %arg, 51
  %tmp1 = and i64 %tmp0, 1023
  %tmp2 = shl i64 %tmp1, 2
  store i64 %tmp2, i64* %ptr
  ret void
}