llvm/test/CodeGen/X86/clear-lowbits.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=X86,X86-NOBMI2
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=X86,X86-NOBMI2
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=X86,X86-NOBMI2
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=X86,X86-BMI2
; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=X86,X86-BMI2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=X64,X64-NOBMI2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=X64,X64-NOBMI2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=X64,X64-NOBMI2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=X64,X64-BMI2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=X64,X64-BMI2

; Patterns:
;    c) x &  (-1 << y)
;   ic) x &  (-1 << (32 - y))
;    d) x >> y << y
;   id) x >> (32 - y) << (32 - y)
; are equivalent, but we prefer the second variant if we have BMI2.

; ---------------------------------------------------------------------------- ;
; Pattern c.
; ---------------------------------------------------------------------------- ;

; 8-bit

define i8 @clear_lowbits8_c0(i8 %val, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_c0:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrb %cl, %al
; X86-NEXT:    shlb %cl, %al
; X86-NEXT:    retl
;
; X64-LABEL: clear_lowbits8_c0:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    shrb %cl, %al
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shlb %cl, %al
; X64-NEXT:    # kill: def $al killed $al killed $eax
; X64-NEXT:    retq
  %mask = shl i8 -1, %numlowbits
  %masked = and i8 %mask, %val
  ret i8 %masked
}

define i8 @clear_lowbits8_c2_load(ptr %w, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_c2_load:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movzbl (%eax), %eax
; X86-NEXT:    shrb %cl, %al
; X86-NEXT:    shlb %cl, %al
; X86-NEXT:    retl
;
; X64-LABEL: clear_lowbits8_c2_load:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    movzbl (%rdi), %eax
; X64-NEXT:    shrb %cl, %al
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shlb %cl, %al
; X64-NEXT:    retq
  %val = load i8, ptr %w
  %mask = shl i8 -1, %numlowbits
  %masked = and i8 %mask, %val
  ret i8 %masked
}

define i8 @clear_lowbits8_c4_commutative(i8 %val, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_c4_commutative:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    shrb %cl, %al
; X86-NEXT:    shlb %cl, %al
; X86-NEXT:    retl
;
; X64-LABEL: clear_lowbits8_c4_commutative:
; X64:       # %bb.0:
; X64-NEXT:    movl %esi, %ecx
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    shrb %cl, %al
; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NEXT:    shlb %cl, %al
; X64-NEXT:    # kill: def $al killed $al killed $eax
; X64-NEXT:    retq
  %mask = shl i8 -1, %numlowbits
  %masked = and i8 %val, %mask ; swapped order
  ret i8 %masked
}

; 16-bit

define i16 @clear_lowbits16_c0(i16 %val, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_c0:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_c0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c0:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movzwl %di, %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_c0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl %di, %eax
; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %mask = shl i16 -1, %numlowbits
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_c1_indexzext(i16 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_c1_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_c1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c1_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movzwl %di, %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_c1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl %di, %eax
; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %sh_prom = zext i8 %numlowbits to i16
  %mask = shl i16 -1, %sh_prom
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_c2_load(ptr %w, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_c2_load:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movzwl (%eax), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_c2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movzwl (%ecx), %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c2_load:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movzwl (%rdi), %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_c2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl (%rdi), %eax
; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %val = load i16, ptr %w
  %mask = shl i16 -1, %numlowbits
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_c3_load_indexzext(ptr %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movzwl (%eax), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movzwl (%ecx), %ecx
; X86-BMI2-NEXT:    shrxl %eax, %ecx, %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movzwl (%rdi), %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl (%rdi), %eax
; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %val = load i16, ptr %w
  %sh_prom = zext i8 %numlowbits to i16
  %mask = shl i16 -1, %sh_prom
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_c4_commutative(i16 %val, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_c4_commutative:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_c4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c4_commutative:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movzwl %di, %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_c4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl %di, %eax
; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %mask = shl i16 -1, %numlowbits
  %masked = and i16 %val, %mask ; swapped order
  ret i16 %masked
}

; 32-bit

define i32 @clear_lowbits32_c0(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_c0:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_c0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c0:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl %edi, %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_c0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %mask = shl i32 -1, %numlowbits
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_c1_indexzext(i32 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_c1_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_c1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c1_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl %edi, %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_c1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %sh_prom = zext i8 %numlowbits to i32
  %mask = shl i32 -1, %sh_prom
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_c2_load(ptr %w, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_c2_load:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movl (%eax), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_c2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, (%eax), %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c2_load:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl (%rdi), %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_c2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, ptr %w
  %mask = shl i32 -1, %numlowbits
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_c3_load_indexzext(ptr %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movl (%eax), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shrxl %ecx, (%eax), %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl (%rdi), %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, ptr %w
  %sh_prom = zext i8 %numlowbits to i32
  %mask = shl i32 -1, %sh_prom
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_c4_commutative(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_c4_commutative:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_c4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c4_commutative:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl %edi, %eax
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_c4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %mask = shl i32 -1, %numlowbits
  %masked = and i32 %val, %mask ; swapped order
  ret i32 %masked
}

; 64-bit

define i64 @clear_lowbits64_c0(i64 %val, i64 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_c0:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB13_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB13_2:
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_c0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB13_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB13_2:
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_c0:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movq %rsi, %rcx
; X64-NOBMI2-NEXT:    movq %rdi, %rax
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_c0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %mask = shl i64 -1, %numlowbits
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_c1_indexzext(i64 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_c1_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB14_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB14_2:
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_c1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB14_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB14_2:
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_c1_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movq %rdi, %rax
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_c1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %sh_prom = zext i8 %numlowbits to i64
  %mask = shl i64 -1, %sh_prom
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_c2_load(ptr %w, i64 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_c2_load:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    pushl %esi
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB15_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB15_2:
; X86-NOBMI2-NEXT:    andl (%esi), %eax
; X86-NOBMI2-NEXT:    andl 4(%esi), %edx
; X86-NOBMI2-NEXT:    popl %esi
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_c2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ebx
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB15_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB15_2:
; X86-BMI2-NEXT:    andl (%ecx), %eax
; X86-BMI2-NEXT:    andl 4(%ecx), %edx
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_c2_load:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movq %rsi, %rcx
; X64-NOBMI2-NEXT:    movq (%rdi), %rax
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_c2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, ptr %w
  %mask = shl i64 -1, %numlowbits
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_c3_load_indexzext(ptr %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_c3_load_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    pushl %esi
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB16_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB16_2:
; X86-NOBMI2-NEXT:    andl (%esi), %eax
; X86-NOBMI2-NEXT:    andl 4(%esi), %edx
; X86-NOBMI2-NEXT:    popl %esi
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_c3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ebx
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB16_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB16_2:
; X86-BMI2-NEXT:    andl (%ecx), %eax
; X86-BMI2-NEXT:    andl 4(%ecx), %edx
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_c3_load_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movq (%rdi), %rax
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_c3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, ptr %w
  %sh_prom = zext i8 %numlowbits to i64
  %mask = shl i64 -1, %sh_prom
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_c4_commutative(i64 %val, i64 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_c4_commutative:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB17_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB17_2:
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_c4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB17_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB17_2:
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_c4_commutative:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movq %rsi, %rcx
; X64-NOBMI2-NEXT:    movq %rdi, %rax
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_c4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %mask = shl i64 -1, %numlowbits
  %masked = and i64 %val, %mask ; swapped order
  ret i64 %masked
}

; ---------------------------------------------------------------------------- ;
; Pattern ic.
; ---------------------------------------------------------------------------- ;

; 8-bit

define i8 @clear_lowbits8_ic0(i8 %val, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_ic0:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movb $8, %cl
; X86-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NEXT:    shrb %cl, %al
; X86-NEXT:    shlb %cl, %al
; X86-NEXT:    retl
;
; X64-LABEL: clear_lowbits8_ic0:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    movb $8, %cl
; X64-NEXT:    subb %sil, %cl
; X64-NEXT:    shrb %cl, %al
; X64-NEXT:    shlb %cl, %al
; X64-NEXT:    # kill: def $al killed $al killed $eax
; X64-NEXT:    retq
  %numhighbits = sub i8 8, %numlowbits
  %mask = shl i8 -1, %numhighbits
  %masked = and i8 %mask, %val
  ret i8 %masked
}

define i8 @clear_lowbits8_ic2_load(ptr %w, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_ic2_load:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movzbl (%eax), %eax
; X86-NEXT:    movb $8, %cl
; X86-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NEXT:    shrb %cl, %al
; X86-NEXT:    shlb %cl, %al
; X86-NEXT:    retl
;
; X64-LABEL: clear_lowbits8_ic2_load:
; X64:       # %bb.0:
; X64-NEXT:    movzbl (%rdi), %eax
; X64-NEXT:    movb $8, %cl
; X64-NEXT:    subb %sil, %cl
; X64-NEXT:    shrb %cl, %al
; X64-NEXT:    shlb %cl, %al
; X64-NEXT:    retq
  %val = load i8, ptr %w
  %numhighbits = sub i8 8, %numlowbits
  %mask = shl i8 -1, %numhighbits
  %masked = and i8 %mask, %val
  ret i8 %masked
}

define i8 @clear_lowbits8_ic4_commutative(i8 %val, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_ic4_commutative:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movb $8, %cl
; X86-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NEXT:    shrb %cl, %al
; X86-NEXT:    shlb %cl, %al
; X86-NEXT:    retl
;
; X64-LABEL: clear_lowbits8_ic4_commutative:
; X64:       # %bb.0:
; X64-NEXT:    movl %edi, %eax
; X64-NEXT:    movb $8, %cl
; X64-NEXT:    subb %sil, %cl
; X64-NEXT:    shrb %cl, %al
; X64-NEXT:    shlb %cl, %al
; X64-NEXT:    # kill: def $al killed $al killed $eax
; X64-NEXT:    retq
  %numhighbits = sub i8 8, %numlowbits
  %mask = shl i8 -1, %numhighbits
  %masked = and i8 %val, %mask ; swapped order
  ret i8 %masked
}

; 16-bit

define i16 @clear_lowbits16_ic0(i16 %val, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic0:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movb $16, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movb $16, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic0:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movzwl %di, %eax
; X64-NOBMI2-NEXT:    movb $16, %cl
; X64-NOBMI2-NEXT:    subb %sil, %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl %di, %eax
; X64-BMI2-NEXT:    movb $16, %cl
; X64-BMI2-NEXT:    subb %sil, %cl
; X64-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i16 16, %numlowbits
  %mask = shl i16 -1, %numhighbits
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_ic1_indexzext(i16 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movb $16, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movb $16, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movzwl %di, %eax
; X64-NOBMI2-NEXT:    movb $16, %cl
; X64-NOBMI2-NEXT:    subb %sil, %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl %di, %eax
; X64-BMI2-NEXT:    movb $16, %cl
; X64-BMI2-NEXT:    subb %sil, %cl
; X64-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i8 16, %numlowbits
  %sh_prom = zext i8 %numhighbits to i16
  %mask = shl i16 -1, %sh_prom
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_ic2_load(ptr %w, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic2_load:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movzwl (%eax), %eax
; X86-NOBMI2-NEXT:    movb $16, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movzwl (%eax), %eax
; X86-BMI2-NEXT:    movb $16, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic2_load:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movzwl (%rdi), %eax
; X64-NOBMI2-NEXT:    movb $16, %cl
; X64-NOBMI2-NEXT:    subb %sil, %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl (%rdi), %eax
; X64-BMI2-NEXT:    movb $16, %cl
; X64-BMI2-NEXT:    subb %sil, %cl
; X64-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %val = load i16, ptr %w
  %numhighbits = sub i16 16, %numlowbits
  %mask = shl i16 -1, %numhighbits
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_ic3_load_indexzext(ptr %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movzwl (%eax), %eax
; X86-NOBMI2-NEXT:    movb $16, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movzwl (%eax), %eax
; X86-BMI2-NEXT:    movb $16, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movzwl (%rdi), %eax
; X64-NOBMI2-NEXT:    movb $16, %cl
; X64-NOBMI2-NEXT:    subb %sil, %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl (%rdi), %eax
; X64-BMI2-NEXT:    movb $16, %cl
; X64-BMI2-NEXT:    subb %sil, %cl
; X64-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %val = load i16, ptr %w
  %numhighbits = sub i8 16, %numlowbits
  %sh_prom = zext i8 %numhighbits to i16
  %mask = shl i16 -1, %sh_prom
  %masked = and i16 %mask, %val
  ret i16 %masked
}

define i16 @clear_lowbits16_ic4_commutative(i16 %val, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic4_commutative:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movb $16, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movb $16, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic4_commutative:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movzwl %di, %eax
; X64-NOBMI2-NEXT:    movb $16, %cl
; X64-NOBMI2-NEXT:    subb %sil, %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movzwl %di, %eax
; X64-BMI2-NEXT:    movb $16, %cl
; X64-BMI2-NEXT:    subb %sil, %cl
; X64-BMI2-NEXT:    shrxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i16 16, %numlowbits
  %mask = shl i16 -1, %numhighbits
  %masked = and i16 %val, %mask ; swapped order
  ret i16 %masked
}

; 32-bit

define i32 @clear_lowbits32_ic0(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic0:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    xorl %ecx, %ecx
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic0:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl %edi, %eax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i32 32, %numlowbits
  %mask = shl i32 -1, %numhighbits
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_ic1_indexzext(i32 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    xorl %ecx, %ecx
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl %edi, %eax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i8 32, %numlowbits
  %sh_prom = zext i8 %numhighbits to i32
  %mask = shl i32 -1, %sh_prom
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_ic2_load(ptr %w, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic2_load:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movl (%eax), %eax
; X86-NOBMI2-NEXT:    xorl %ecx, %ecx
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, (%eax), %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic2_load:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl (%rdi), %eax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, ptr %w
  %numhighbits = sub i32 32, %numlowbits
  %mask = shl i32 -1, %numhighbits
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_ic3_load_indexzext(ptr %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    movl (%eax), %eax
; X86-NOBMI2-NEXT:    xorl %ecx, %ecx
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    xorl %ecx, %ecx
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    shrxl %ecx, (%eax), %eax
; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl (%rdi), %eax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %val = load i32, ptr %w
  %numhighbits = sub i8 32, %numlowbits
  %sh_prom = zext i8 %numhighbits to i32
  %mask = shl i32 -1, %sh_prom
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i32 @clear_lowbits32_ic4_commutative(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic4_commutative:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    xorl %ecx, %ecx
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    shrl %cl, %eax
; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic4_commutative:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl %edi, %eax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrl %cl, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i32 32, %numlowbits
  %mask = shl i32 -1, %numhighbits
  %masked = and i32 %val, %mask ; swapped order
  ret i32 %masked
}

; 64-bit

define i64 @clear_lowbits64_ic0(i64 %val, i64 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_ic0:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movb $64, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB31_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB31_2:
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_ic0:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB31_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB31_2:
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_ic0:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movq %rsi, %rcx
; X64-NOBMI2-NEXT:    movq %rdi, %rax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic0:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i64 64, %numlowbits
  %mask = shl i64 -1, %numhighbits
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_ic1_indexzext(i64 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_ic1_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movb $64, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB32_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB32_2:
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_ic1_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB32_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB32_2:
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_ic1_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movq %rdi, %rax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic1_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i8 64, %numlowbits
  %sh_prom = zext i8 %numhighbits to i64
  %mask = shl i64 -1, %sh_prom
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_ic2_load(ptr %w, i64 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_ic2_load:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    pushl %esi
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI2-NEXT:    movb $64, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB33_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB33_2:
; X86-NOBMI2-NEXT:    andl (%esi), %eax
; X86-NOBMI2-NEXT:    andl 4(%esi), %edx
; X86-NOBMI2-NEXT:    popl %esi
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_ic2_load:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb $64, %bl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB33_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB33_2:
; X86-BMI2-NEXT:    andl (%ecx), %eax
; X86-BMI2-NEXT:    andl 4(%ecx), %edx
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_ic2_load:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movq %rsi, %rcx
; X64-NOBMI2-NEXT:    movq (%rdi), %rax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic2_load:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, ptr %w
  %numhighbits = sub i64 64, %numlowbits
  %mask = shl i64 -1, %numhighbits
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_ic3_load_indexzext(ptr %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    pushl %esi
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI2-NEXT:    movb $64, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB34_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB34_2:
; X86-NOBMI2-NEXT:    andl (%esi), %eax
; X86-NOBMI2-NEXT:    andl 4(%esi), %edx
; X86-NOBMI2-NEXT:    popl %esi
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movb $64, %bl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %bl
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    je .LBB34_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB34_2:
; X86-BMI2-NEXT:    andl (%ecx), %eax
; X86-BMI2-NEXT:    andl 4(%ecx), %edx
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movq (%rdi), %rax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %val = load i64, ptr %w
  %numhighbits = sub i8 64, %numlowbits
  %sh_prom = zext i8 %numhighbits to i64
  %mask = shl i64 -1, %sh_prom
  %masked = and i64 %mask, %val
  ret i64 %masked
}

define i64 @clear_lowbits64_ic4_commutative(i64 %val, i64 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits64_ic4_commutative:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movb $64, %cl
; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    je .LBB35_2
; X86-NOBMI2-NEXT:  # %bb.1:
; X86-NOBMI2-NEXT:    movl %eax, %edx
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:  .LBB35_2:
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: clear_lowbits64_ic4_commutative:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movb $64, %cl
; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
; X86-BMI2-NEXT:    testb $32, %cl
; X86-BMI2-NEXT:    je .LBB35_2
; X86-BMI2-NEXT:  # %bb.1:
; X86-BMI2-NEXT:    movl %eax, %edx
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:  .LBB35_2:
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: clear_lowbits64_ic4_commutative:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movq %rsi, %rcx
; X64-NOBMI2-NEXT:    movq %rdi, %rax
; X64-NOBMI2-NEXT:    negb %cl
; X64-NOBMI2-NEXT:    shrq %cl, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic4_commutative:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    negb %sil
; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    retq
  %numhighbits = sub i64 64, %numlowbits
  %mask = shl i64 -1, %numhighbits
  %masked = and i64 %val, %mask ; swapped order
  ret i64 %masked
}

; ---------------------------------------------------------------------------- ;
; Multi-use tests
; ---------------------------------------------------------------------------- ;

define i32 @oneuse32_c(i32 %val, i32 %numlowbits, ptr %escape) nounwind {
; X86-NOBMI2-LABEL: oneuse32_c:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl $-1, %eax
; X86-NOBMI2-NEXT:    shll %cl, %eax
; X86-NOBMI2-NEXT:    movl %eax, (%edx)
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: oneuse32_c:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
; X86-BMI2-NEXT:    movl %eax, (%ecx)
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: oneuse32_c:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movl %esi, %ecx
; X64-NOBMI2-NEXT:    movl $-1, %eax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT:    shll %cl, %eax
; X64-NOBMI2-NEXT:    movl %eax, (%rdx)
; X64-NOBMI2-NEXT:    andl %edi, %eax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: oneuse32_c:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movl $-1, %eax
; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
; X64-BMI2-NEXT:    movl %eax, (%rdx)
; X64-BMI2-NEXT:    andl %edi, %eax
; X64-BMI2-NEXT:    retq
  %mask = shl i32 -1, %numlowbits
  store i32 %mask, ptr %escape
  %masked = and i32 %mask, %val
  ret i32 %masked
}

define i64 @oneuse64(i64 %val, i64 %numlowbits, ptr %escape) nounwind {
; X86-NOBMI2-LABEL: oneuse64:
; X86-NOBMI2:       # %bb.0:
; X86-NOBMI2-NEXT:    pushl %edi
; X86-NOBMI2-NEXT:    pushl %esi
; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
; X86-NOBMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
; X86-NOBMI2-NEXT:    movl $-1, %edx
; X86-NOBMI2-NEXT:    movl $-1, %edi
; X86-NOBMI2-NEXT:    shll %cl, %edi
; X86-NOBMI2-NEXT:    xorl %eax, %eax
; X86-NOBMI2-NEXT:    testb $32, %cl
; X86-NOBMI2-NEXT:    jne .LBB37_1
; X86-NOBMI2-NEXT:  # %bb.2:
; X86-NOBMI2-NEXT:    movl %edi, %eax
; X86-NOBMI2-NEXT:    jmp .LBB37_3
; X86-NOBMI2-NEXT:  .LBB37_1:
; X86-NOBMI2-NEXT:    movl %edi, %edx
; X86-NOBMI2-NEXT:  .LBB37_3:
; X86-NOBMI2-NEXT:    movl %edx, 4(%esi)
; X86-NOBMI2-NEXT:    movl %eax, (%esi)
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT:    popl %esi
; X86-NOBMI2-NEXT:    popl %edi
; X86-NOBMI2-NEXT:    retl
;
; X86-BMI2-LABEL: oneuse64:
; X86-BMI2:       # %bb.0:
; X86-BMI2-NEXT:    pushl %ebx
; X86-BMI2-NEXT:    pushl %esi
; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT:    movzbl {{[0-9]+}}(%esp), %ebx
; X86-BMI2-NEXT:    movl $-1, %edx
; X86-BMI2-NEXT:    shlxl %ebx, %edx, %esi
; X86-BMI2-NEXT:    xorl %eax, %eax
; X86-BMI2-NEXT:    testb $32, %bl
; X86-BMI2-NEXT:    jne .LBB37_1
; X86-BMI2-NEXT:  # %bb.2:
; X86-BMI2-NEXT:    movl %esi, %eax
; X86-BMI2-NEXT:    jmp .LBB37_3
; X86-BMI2-NEXT:  .LBB37_1:
; X86-BMI2-NEXT:    movl %esi, %edx
; X86-BMI2-NEXT:  .LBB37_3:
; X86-BMI2-NEXT:    movl %edx, 4(%ecx)
; X86-BMI2-NEXT:    movl %eax, (%ecx)
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edx
; X86-BMI2-NEXT:    popl %esi
; X86-BMI2-NEXT:    popl %ebx
; X86-BMI2-NEXT:    retl
;
; X64-NOBMI2-LABEL: oneuse64:
; X64-NOBMI2:       # %bb.0:
; X64-NOBMI2-NEXT:    movq %rsi, %rcx
; X64-NOBMI2-NEXT:    movq $-1, %rax
; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT:    shlq %cl, %rax
; X64-NOBMI2-NEXT:    movq %rax, (%rdx)
; X64-NOBMI2-NEXT:    andq %rdi, %rax
; X64-NOBMI2-NEXT:    retq
;
; X64-BMI2-LABEL: oneuse64:
; X64-BMI2:       # %bb.0:
; X64-BMI2-NEXT:    movq $-1, %rax
; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
; X64-BMI2-NEXT:    movq %rax, (%rdx)
; X64-BMI2-NEXT:    andq %rdi, %rax
; X64-BMI2-NEXT:    retq
  %mask = shl i64 -1, %numlowbits
  store i64 %mask, ptr %escape
  %masked = and i64 %mask, %val
  ret i64 %masked
}