[X86] Add tests for low/high bit clearing with different attributes.

D48768 may turn some of these into shifts.

Reviewers: spatel

Reviewed By: spatel

Subscribers: spatel, RKSimon, llvm-commits, craig.topper

Differential Revision: https://reviews.llvm.org/D48767

llvm-svn: 336224

diff --git a/llvm/test/CodeGen/X86/clear-highbits.ll b/llvm/test/CodeGen/X86/clear-highbits.ll
new file mode 100644 (file)
index 0000000..9e5f32a
--- /dev/null
+++ b/llvm/test/CodeGen/X86/clear-highbits.ll
@@ -0,0 +1,956 @@
+; 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=CHECK,X86,NOBMI2,X86-NOBMI2,FALLBACK0,X86-FALLBACK0
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI2,X86-NOBMI2,FALLBACK1,X86-FALLBACK1
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI2,X86-NOBMI2,FALLBACK2,X86-FALLBACK2
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI2,X86-BMI2,FALLBACK3,X86-FALLBACK3
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI2,X86-BMI2,FALLBACK4,X86-FALLBACK4
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI2,X64-NOBMI2,FALLBACK0,X64-FALLBACK0
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI2,X64-NOBMI2,FALLBACK1,X64-FALLBACK1
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI2,X64-NOBMI2,FALLBACK2,X64-FALLBACK2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI2,X64-BMI2,FALLBACK3,X64-FALLBACK3
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI2,X64-BMI2,FALLBACK4,X64-FALLBACK4
+
+; Patterns:
+;    c) x &  (-1 >> y)
+;    d) x << y >> y
+; are equivalent, but we prefer the second variant if we have BMI2.
+
+; We do not test the variant where y = (32 - z), because that is BMI2's BZHI.
+
+; ---------------------------------------------------------------------------- ;
+; 8-bit
+; ---------------------------------------------------------------------------- ;
+
+define i8 @clear_highbits8_c0(i8 %val, i8 %numhighbits) nounwind {
+; X86-LABEL: clear_highbits8_c0:
+; X86:       # %bb.0:
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shrb %cl, %al
+; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_highbits8_c0:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    movl %esi, %ecx
+; X64-NEXT:    shrb %cl, %al
+; X64-NEXT:    andb %dil, %al
+; X64-NEXT:    retq
+  %mask = lshr i8 -1, %numhighbits
+  %masked = and i8 %mask, %val
+  ret i8 %masked
+}
+
+define i8 @clear_highbits8_c2_load(i8* %w, i8 %numhighbits) nounwind {
+; X86-LABEL: clear_highbits8_c2_load:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shrb %cl, %al
+; X86-NEXT:    andb (%edx), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_highbits8_c2_load:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    movl %esi, %ecx
+; X64-NEXT:    shrb %cl, %al
+; X64-NEXT:    andb (%rdi), %al
+; X64-NEXT:    retq
+  %val = load i8, i8* %w
+  %mask = lshr i8 -1, %numhighbits
+  %masked = and i8 %mask, %val
+  ret i8 %masked
+}
+
+define i8 @clear_highbits8_c4_commutative(i8 %val, i8 %numhighbits) nounwind {
+; X86-LABEL: clear_highbits8_c4_commutative:
+; X86:       # %bb.0:
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shrb %cl, %al
+; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_highbits8_c4_commutative:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    movl %esi, %ecx
+; X64-NEXT:    shrb %cl, %al
+; X64-NEXT:    andb %dil, %al
+; X64-NEXT:    retq
+  %mask = lshr i8 -1, %numhighbits
+  %masked = and i8 %val, %mask ; swapped order
+  ret i8 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; 16-bit
+; ---------------------------------------------------------------------------- ;
+
+define i16 @clear_highbits16_c0(i16 %val, i16 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits16_c0:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits16_c0:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $65535, %ecx # imm = 0xFFFF
+; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits16_c0:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits16_c0:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %eax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i16 -1, %numhighbits
+  %masked = and i16 %mask, %val
+  ret i16 %masked
+}
+
+define i16 @clear_highbits16_c1_indexzext(i16 %val, i8 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits16_c1_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits16_c1_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $65535, %ecx # imm = 0xFFFF
+; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits16_c1_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits16_c1_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %eax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %sh_prom = zext i8 %numhighbits to i16
+  %mask = lshr i16 -1, %sh_prom
+  %masked = and i16 %mask, %val
+  ret i16 %masked
+}
+
+define i16 @clear_highbits16_c2_load(i16* %w, i16 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits16_c2_load:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andw (%edx), %ax
+; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits16_c2_load:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; X86-BMI2-NEXT:    shrxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andw (%ecx), %ax
+; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits16_c2_load:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andw (%rdi), %ax
+; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits16_c2_load:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andw (%rdi), %ax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %val = load i16, i16* %w
+  %mask = lshr i16 -1, %numhighbits
+  %masked = and i16 %mask, %val
+  ret i16 %masked
+}
+
+define i16 @clear_highbits16_c3_load_indexzext(i16* %w, i8 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits16_c3_load_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andw (%edx), %ax
+; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits16_c3_load_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $65535, %edx # imm = 0xFFFF
+; X86-BMI2-NEXT:    shrxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andw (%ecx), %ax
+; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits16_c3_load_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andw (%rdi), %ax
+; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits16_c3_load_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andw (%rdi), %ax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %val = load i16, i16* %w
+  %sh_prom = zext i8 %numhighbits to i16
+  %mask = lshr i16 -1, %sh_prom
+  %masked = and i16 %mask, %val
+  ret i16 %masked
+}
+
+define i16 @clear_highbits16_c4_commutative(i16 %val, i16 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits16_c4_commutative:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; X86-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits16_c4_commutative:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $65535, %ecx # imm = 0xFFFF
+; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits16_c4_commutative:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits16_c4_commutative:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %eax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i16 -1, %numhighbits
+  %masked = and i16 %val, %mask ; swapped order
+  ret i16 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; 32-bit
+; ---------------------------------------------------------------------------- ;
+
+define i32 @clear_highbits32_c0(i32 %val, i32 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits32_c0:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits32_c0:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits32_c0:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits32_c0:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %mask, %val
+  ret i32 %masked
+}
+
+define i32 @clear_highbits32_c1_indexzext(i32 %val, i8 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits32_c1_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits32_c1_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits32_c1_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits32_c1_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %sh_prom = zext i8 %numhighbits to i32
+  %mask = lshr i32 -1, %sh_prom
+  %masked = and i32 %mask, %val
+  ret i32 %masked
+}
+
+define i32 @clear_highbits32_c2_load(i32* %w, i32 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits32_c2_load:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andl (%edx), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits32_c2_load:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shrxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andl (%ecx), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits32_c2_load:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl (%rdi), %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits32_c2_load:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl (%rdi), %eax
+; X64-BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %mask, %val
+  ret i32 %masked
+}
+
+define i32 @clear_highbits32_c3_load_indexzext(i32* %w, i8 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits32_c3_load_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andl (%edx), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits32_c3_load_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shrxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andl (%ecx), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits32_c3_load_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl (%rdi), %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits32_c3_load_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl (%rdi), %eax
+; X64-BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %sh_prom = zext i8 %numhighbits to i32
+  %mask = lshr i32 -1, %sh_prom
+  %masked = and i32 %mask, %val
+  ret i32 %masked
+}
+
+define i32 @clear_highbits32_c4_commutative(i32 %val, i32 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits32_c4_commutative:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shrl %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_highbits32_c4_commutative:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shrxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits32_c4_commutative:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits32_c4_commutative:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %eax
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %val, %mask ; swapped order
+  ret i32 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; 64-bit
+; ---------------------------------------------------------------------------- ;
+
+define i64 @clear_highbits64_c0(i64 %val, i64 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits64_c0:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    movl $-1, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB13_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c0:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB13_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c0:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits64_c0:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shrxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %mask, %val
+  ret i64 %masked
+}
+
+define i64 @clear_highbits64_c1_indexzext(i64 %val, i8 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits64_c1_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    movl $-1, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB14_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c1_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB14_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c1_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits64_c1_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shrxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %sh_prom = zext i8 %numhighbits to i64
+  %mask = lshr i64 -1, %sh_prom
+  %masked = and i64 %mask, %val
+  ret i64 %masked
+}
+
+define i64 @clear_highbits64_c2_load(i64* %w, i64 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits64_c2_load:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    movl $-1, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB15_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c2_load:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB15_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI2-NEXT:  .LBB15_2:
+; X86-BMI2-NEXT:    andl (%esi), %eax
+; X86-BMI2-NEXT:    andl 4(%esi), %edx
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits64_c2_load:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    andq (%rdi), %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits64_c2_load:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shrxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq (%rdi), %rax
+; X64-BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %mask, %val
+  ret i64 %masked
+}
+
+define i64 @clear_highbits64_c3_load_indexzext(i64* %w, i8 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits64_c3_load_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    movl $-1, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB16_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c3_load_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB16_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI2-NEXT:  .LBB16_2:
+; X86-BMI2-NEXT:    andl (%esi), %eax
+; X86-BMI2-NEXT:    andl 4(%esi), %edx
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_highbits64_c3_load_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    andq (%rdi), %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits64_c3_load_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shrxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq (%rdi), %rax
+; X64-BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %sh_prom = zext i8 %numhighbits to i64
+  %mask = lshr i64 -1, %sh_prom
+  %masked = and i64 %mask, %val
+  ret i64 %masked
+}
+
+define i64 @clear_highbits64_c4_commutative(i64 %val, i64 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: clear_highbits64_c4_commutative:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    movl $-1, %edx
+; X86-NOBMI2-NEXT:    shrl %cl, %edx
+; X86-NOBMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB17_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edx, %eax
+; X86-NOBMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c4_commutative:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %eax
+; X86-BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB17_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edx, %eax
+; X86-BMI2-NEXT:    xorl %edx, %edx
+; 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_highbits64_c4_commutative:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_highbits64_c4_commutative:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shrxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq %rdi, %rax
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %val, %mask ; swapped order
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Multi-use tests
+; ---------------------------------------------------------------------------- ;
+
+declare void @use32(i32)
+declare void @use64(i64)
+
+define i32 @oneuse32(i32 %val, i32 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: oneuse32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    subl $8, %esp
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %esi
+; X86-NOBMI2-NEXT:    shrl %cl, %esi
+; X86-NOBMI2-NEXT:    movl %esi, (%esp)
+; X86-NOBMI2-NEXT:    calll use32
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movl %esi, %eax
+; X86-NOBMI2-NEXT:    addl $8, %esp
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: oneuse32:
+; 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 $-1, %ecx
+; X86-BMI2-NEXT:    shrxl %eax, %ecx, %esi
+; X86-BMI2-NEXT:    movl %esi, (%esp)
+; X86-BMI2-NEXT:    calll use32
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    movl %esi, %eax
+; X86-BMI2-NEXT:    addl $8, %esp
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: oneuse32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    pushq %rbp
+; X64-NOBMI2-NEXT:    pushq %rbx
+; X64-NOBMI2-NEXT:    pushq %rax
+; X64-NOBMI2-NEXT:    movl %edi, %ebx
+; X64-NOBMI2-NEXT:    movl $-1, %ebp
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrl %cl, %ebp
+; X64-NOBMI2-NEXT:    movl %ebp, %edi
+; X64-NOBMI2-NEXT:    callq use32
+; X64-NOBMI2-NEXT:    andl %ebx, %ebp
+; X64-NOBMI2-NEXT:    movl %ebp, %eax
+; X64-NOBMI2-NEXT:    addq $8, %rsp
+; X64-NOBMI2-NEXT:    popq %rbx
+; X64-NOBMI2-NEXT:    popq %rbp
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: oneuse32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    pushq %rbp
+; X64-BMI2-NEXT:    pushq %rbx
+; X64-BMI2-NEXT:    pushq %rax
+; X64-BMI2-NEXT:    movl %edi, %ebx
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shrxl %esi, %eax, %ebp
+; X64-BMI2-NEXT:    movl %ebp, %edi
+; X64-BMI2-NEXT:    callq use32
+; X64-BMI2-NEXT:    andl %ebx, %ebp
+; X64-BMI2-NEXT:    movl %ebp, %eax
+; X64-BMI2-NEXT:    addq $8, %rsp
+; X64-BMI2-NEXT:    popq %rbx
+; X64-BMI2-NEXT:    popq %rbp
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i32 -1, %numhighbits
+  call void @use32(i32 %mask)
+  %masked = and i32 %mask, %val
+  ret i32 %masked
+}
+
+define i64 @oneuse64(i64 %val, i64 %numhighbits) nounwind {
+; X86-NOBMI2-LABEL: oneuse64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %edi
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    pushl %eax
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %esi
+; X86-NOBMI2-NEXT:    movl $-1, %edi
+; X86-NOBMI2-NEXT:    shrl %cl, %edi
+; X86-NOBMI2-NEXT:    shrdl %cl, %esi, %esi
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB19_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edi, %esi
+; X86-NOBMI2-NEXT:    xorl %edi, %edi
+; X86-NOBMI2-NEXT:  .LBB19_2:
+; X86-NOBMI2-NEXT:    subl $8, %esp
+; X86-NOBMI2-NEXT:    pushl %edi
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    calll use64
+; X86-NOBMI2-NEXT:    addl $16, %esp
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edi
+; X86-NOBMI2-NEXT:    movl %esi, %eax
+; X86-NOBMI2-NEXT:    movl %edi, %edx
+; X86-NOBMI2-NEXT:    addl $4, %esp
+; 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 %edi
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    pushl %eax
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %esi
+; X86-BMI2-NEXT:    shrxl %ecx, %esi, %edi
+; X86-BMI2-NEXT:    shrdl %cl, %esi, %esi
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB19_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edi, %esi
+; X86-BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI2-NEXT:  .LBB19_2:
+; X86-BMI2-NEXT:    subl $8, %esp
+; X86-BMI2-NEXT:    pushl %edi
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    calll use64
+; X86-BMI2-NEXT:    addl $16, %esp
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edi
+; X86-BMI2-NEXT:    movl %esi, %eax
+; X86-BMI2-NEXT:    movl %edi, %edx
+; X86-BMI2-NEXT:    addl $4, %esp
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    popl %edi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: oneuse64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    pushq %r14
+; X64-NOBMI2-NEXT:    pushq %rbx
+; X64-NOBMI2-NEXT:    pushq %rax
+; X64-NOBMI2-NEXT:    movq %rdi, %r14
+; X64-NOBMI2-NEXT:    movq $-1, %rbx
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shrq %cl, %rbx
+; X64-NOBMI2-NEXT:    movq %rbx, %rdi
+; X64-NOBMI2-NEXT:    callq use64
+; X64-NOBMI2-NEXT:    andq %r14, %rbx
+; X64-NOBMI2-NEXT:    movq %rbx, %rax
+; X64-NOBMI2-NEXT:    addq $8, %rsp
+; X64-NOBMI2-NEXT:    popq %rbx
+; X64-NOBMI2-NEXT:    popq %r14
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: oneuse64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    pushq %r14
+; X64-BMI2-NEXT:    pushq %rbx
+; X64-BMI2-NEXT:    pushq %rax
+; X64-BMI2-NEXT:    movq %rdi, %r14
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shrxq %rsi, %rax, %rbx
+; X64-BMI2-NEXT:    movq %rbx, %rdi
+; X64-BMI2-NEXT:    callq use64
+; X64-BMI2-NEXT:    andq %r14, %rbx
+; X64-BMI2-NEXT:    movq %rbx, %rax
+; X64-BMI2-NEXT:    addq $8, %rsp
+; X64-BMI2-NEXT:    popq %rbx
+; X64-BMI2-NEXT:    popq %r14
+; X64-BMI2-NEXT:    retq
+  %mask = lshr i64 -1, %numhighbits
+  call void @use64(i64 %mask)
+  %masked = and i64 %mask, %val
+  ret i64 %masked
+}

diff --git a/llvm/test/CodeGen/X86/clear-lowbits.ll b/llvm/test/CodeGen/X86/clear-lowbits.ll
new file mode 100644 (file)
index 0000000..b05478f
--- /dev/null
+++ b/llvm/test/CodeGen/X86/clear-lowbits.ll
@@ -0,0 +1,1820 @@
+; 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=CHECK,X86,NOBMI2,X86-NOBMI2,FALLBACK0,X86-FALLBACK0
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI2,X86-NOBMI2,FALLBACK1,X86-FALLBACK1
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI2,X86-NOBMI2,FALLBACK2,X86-FALLBACK2
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI2,X86-BMI2,FALLBACK3,X86-FALLBACK3
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI2,X86-BMI2,FALLBACK4,X86-FALLBACK4
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI2,X64-NOBMI2,FALLBACK0,X64-FALLBACK0
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI2,X64-NOBMI2,FALLBACK1,X64-FALLBACK1
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI2,X64-NOBMI2,FALLBACK2,X64-FALLBACK2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI2,X64-BMI2,FALLBACK3,X64-FALLBACK3
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI2,X64-BMI2,FALLBACK4,X64-FALLBACK4
+
+; 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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shlb %cl, %al
+; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_lowbits8_c0:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    movl %esi, %ecx
+; X64-NEXT:    shlb %cl, %al
+; X64-NEXT:    andb %dil, %al
+; X64-NEXT:    retq
+  %mask = shl i8 -1, %numlowbits
+  %masked = and i8 %mask, %val
+  ret i8 %masked
+}
+
+define i8 @clear_lowbits8_c2_load(i8* %w, i8 %numlowbits) nounwind {
+; X86-LABEL: clear_lowbits8_c2_load:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shlb %cl, %al
+; X86-NEXT:    andb (%edx), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_lowbits8_c2_load:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    movl %esi, %ecx
+; X64-NEXT:    shlb %cl, %al
+; X64-NEXT:    andb (%rdi), %al
+; X64-NEXT:    retq
+  %val = load i8, i8* %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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shlb %cl, %al
+; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_lowbits8_c4_commutative:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    movl %esi, %ecx
+; X64-NEXT:    shlb %cl, %al
+; X64-NEXT:    andb %dil, %al
+; 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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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 $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %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:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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 $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %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:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %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(i16* %w, i16 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits16_c2_load:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw (%edx), %ax
+; 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:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andw (%ecx), %ax
+; 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 $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andw (%rdi), %ax
+; 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:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andw (%rdi), %ax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %val = load i16, i16* %w
+  %mask = shl i16 -1, %numlowbits
+  %masked = and i16 %mask, %val
+  ret i16 %masked
+}
+
+define i16 @clear_lowbits16_c3_load_indexzext(i16* %w, i8 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits16_c3_load_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw (%edx), %ax
+; 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:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andw (%ecx), %ax
+; 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 $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andw (%rdi), %ax
+; 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:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andw (%rdi), %ax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %val = load i16, i16* %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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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 $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %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:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_c0:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_c0:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_c0:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_c1_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_c1_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_c1_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %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(i32* %w, i32 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits32_c2_load:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl (%edx), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_c2_load:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andl (%ecx), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_c2_load:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl (%rdi), %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_c2_load:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl (%rdi), %eax
+; X64-BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %mask = shl i32 -1, %numlowbits
+  %masked = and i32 %mask, %val
+  ret i32 %masked
+}
+
+define i32 @clear_lowbits32_c3_load_indexzext(i32* %w, i8 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits32_c3_load_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl (%edx), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_c3_load_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andl (%ecx), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_c3_load_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl (%rdi), %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_c3_load_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl (%rdi), %eax
+; X64-BMI2-NEXT:    retq
+  %val = load i32, i32* %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:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_c4_commutative:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_c4_commutative:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_c4_commutative:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT:    andl %edi, %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:    movb {{[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:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits64_c0:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_c0:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_c0:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq %rdi, %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:    movb {{[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:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits64_c1_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_c1_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %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:    movq $-1, %rax
+; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq %rdi, %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(i64* %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:    movb {{[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:    shldl %cl, %edx, %edx
+; 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 4(%esi), %edx
+; X86-NOBMI2-NEXT:    andl (%esi), %eax
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits64_c2_load:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; X86-BMI2-NEXT:    testb $32, %cl
+; 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 4(%esi), %edx
+; X86-BMI2-NEXT:    andl (%esi), %eax
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_c2_load:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq (%rdi), %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_c2_load:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq (%rdi), %rax
+; X64-BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %mask = shl i64 -1, %numlowbits
+  %masked = and i64 %mask, %val
+  ret i64 %masked
+}
+
+define i64 @clear_lowbits64_c3_load_indexzext(i64* %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:    movb {{[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:    shldl %cl, %edx, %edx
+; 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 4(%esi), %edx
+; X86-NOBMI2-NEXT:    andl (%esi), %eax
+; 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 %esi
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; X86-BMI2-NEXT:    testb $32, %cl
+; 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 4(%esi), %edx
+; X86-BMI2-NEXT:    andl (%esi), %eax
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_c3_load_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq (%rdi), %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:    movq $-1, %rax
+; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq (%rdi), %rax
+; X64-BMI2-NEXT:    retq
+  %val = load i64, i64* %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:    movb {{[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:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits64_c4_commutative:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_c4_commutative:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_c4_commutative:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rax
+; X64-BMI2-NEXT:    andq %rdi, %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:    movb $8, %cl
+; X86-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shlb %cl, %al
+; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_lowbits8_ic0:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $8, %cl
+; X64-NEXT:    subb %sil, %cl
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    shlb %cl, %al
+; X64-NEXT:    andb %dil, %al
+; 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(i8* %w, i8 %numlowbits) nounwind {
+; X86-LABEL: clear_lowbits8_ic2_load:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movb $8, %cl
+; X86-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shlb %cl, %al
+; X86-NEXT:    andb (%edx), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_lowbits8_ic2_load:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $8, %cl
+; X64-NEXT:    subb %sil, %cl
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    shlb %cl, %al
+; X64-NEXT:    andb (%rdi), %al
+; X64-NEXT:    retq
+  %val = load i8, i8* %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:    movb $8, %cl
+; X86-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NEXT:    movb $-1, %al
+; X86-NEXT:    shlb %cl, %al
+; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; X86-NEXT:    retl
+;
+; X64-LABEL: clear_lowbits8_ic4_commutative:
+; X64:       # %bb.0:
+; X64-NEXT:    movb $8, %cl
+; X64-NEXT:    subb %sil, %cl
+; X64-NEXT:    movb $-1, %al
+; X64-NEXT:    shlb %cl, %al
+; X64-NEXT:    andb %dil, %al
+; 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:    movw $16, %cx
+; X86-NOBMI2-NEXT:    subw {{[0-9]+}}(%esp), %cx
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $cx
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movw $16, %ax
+; X86-BMI2-NEXT:    subw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movl $16, %ecx
+; X64-NOBMI2-NEXT:    subl %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:    andl %edi, %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:    movl $16, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl %edi, %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:    movb $16, %cl
+; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movb $16, %al
+; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movb $16, %cl
+; X64-NOBMI2-NEXT:    subb %sil, %cl
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %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:    movb $16, %al
+; X64-BMI2-NEXT:    subb %sil, %al
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl %edi, %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(i16* %w, i16 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits16_ic2_load:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movw $16, %cx
+; X86-NOBMI2-NEXT:    subw {{[0-9]+}}(%esp), %cx
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $cx
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw (%edx), %ax
+; 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), %ecx
+; X86-BMI2-NEXT:    movw $16, %ax
+; X86-BMI2-NEXT:    subw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andw (%ecx), %ax
+; 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:    movl $16, %ecx
+; X64-NOBMI2-NEXT:    subl %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:    andw (%rdi), %ax
+; 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:    movl $16, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andw (%rdi), %ax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %val = load i16, i16* %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(i16* %w, i8 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb $16, %cl
+; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw (%edx), %ax
+; 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), %ecx
+; X86-BMI2-NEXT:    movb $16, %al
+; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andw (%ecx), %ax
+; 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:    movb $16, %cl
+; X64-NOBMI2-NEXT:    subb %sil, %cl
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andw (%rdi), %ax
+; 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:    movb $16, %al
+; X64-BMI2-NEXT:    subb %sil, %al
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andw (%rdi), %ax
+; X64-BMI2-NEXT:    # kill: def $ax killed $ax killed $eax
+; X64-BMI2-NEXT:    retq
+  %val = load i16, i16* %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:    movw $16, %cx
+; X86-NOBMI2-NEXT:    subw {{[0-9]+}}(%esp), %cx
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $cx
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movw $16, %ax
+; X86-BMI2-NEXT:    subw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andw {{[0-9]+}}(%esp), %ax
+; 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:    movl $16, %ecx
+; X64-NOBMI2-NEXT:    subl %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:    andl %edi, %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:    movl $16, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl %edi, %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 $32, %ecx
+; X86-NOBMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_ic0:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl $32, %eax
+; X86-BMI2-NEXT:    subl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_ic0:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $32, %ecx
+; X64-NOBMI2-NEXT:    subl %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:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_ic0:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $32, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl %edi, %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:    movb $32, %cl
+; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_ic1_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movb $32, %al
+; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_ic1_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movb $32, %cl
+; X64-NOBMI2-NEXT:    subb %sil, %cl
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_ic1_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movb $32, %al
+; X64-BMI2-NEXT:    subb %sil, %al
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl %edi, %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(i32* %w, i32 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits32_ic2_load:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movl $32, %ecx
+; X86-NOBMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl (%edx), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_ic2_load:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movl $32, %eax
+; X86-BMI2-NEXT:    subl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andl (%ecx), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_ic2_load:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $32, %ecx
+; X64-NOBMI2-NEXT:    subl %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:    andl (%rdi), %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_ic2_load:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $32, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl (%rdi), %eax
+; X64-BMI2-NEXT:    retq
+  %val = load i32, i32* %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(i32* %w, i8 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT:    movb $32, %cl
+; X86-NOBMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl (%edx), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movb $32, %al
+; X86-BMI2-NEXT:    subb {{[0-9]+}}(%esp), %al
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %eax, %edx, %eax
+; X86-BMI2-NEXT:    andl (%ecx), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movb $32, %cl
+; X64-NOBMI2-NEXT:    subb %sil, %cl
+; X64-NOBMI2-NEXT:    movl $-1, %eax
+; X64-NOBMI2-NEXT:    shll %cl, %eax
+; X64-NOBMI2-NEXT:    andl (%rdi), %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movb $32, %al
+; X64-BMI2-NEXT:    subb %sil, %al
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl (%rdi), %eax
+; X64-BMI2-NEXT:    retq
+  %val = load i32, i32* %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 $32, %ecx
+; X86-NOBMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI2-NEXT:    movl $-1, %eax
+; X86-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI2-NEXT:    shll %cl, %eax
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits32_ic4_commutative:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl $32, %eax
+; X86-BMI2-NEXT:    subl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    movl $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits32_ic4_commutative:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $32, %ecx
+; X64-NOBMI2-NEXT:    subl %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:    andl %edi, %eax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits32_ic4_commutative:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $32, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movl $-1, %ecx
+; X64-BMI2-NEXT:    shlxl %eax, %ecx, %eax
+; X64-BMI2-NEXT:    andl %edi, %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:    movl $64, %ecx
+; X86-NOBMI2-NEXT:    subl {{[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:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits64_ic0:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl $64, %ecx
+; X86-BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_ic0:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $64, %ecx
+; X64-NOBMI2-NEXT:    subl %esi, %ecx
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_ic0:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $64, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movq $-1, %rcx
+; X64-BMI2-NEXT:    shlxq %rax, %rcx, %rax
+; X64-BMI2-NEXT:    andq %rdi, %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:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; 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:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_ic1_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movb $64, %cl
+; X64-NOBMI2-NEXT:    subb %sil, %cl
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_ic1_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movb $64, %al
+; X64-BMI2-NEXT:    subb %sil, %al
+; X64-BMI2-NEXT:    movq $-1, %rcx
+; X64-BMI2-NEXT:    shlxq %rax, %rcx, %rax
+; X64-BMI2-NEXT:    andq %rdi, %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(i64* %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:    movl $64, %ecx
+; X86-NOBMI2-NEXT:    subl {{[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:    shldl %cl, %edx, %edx
+; 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 4(%esi), %edx
+; X86-NOBMI2-NEXT:    andl (%esi), %eax
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits64_ic2_load:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    movl $64, %ecx
+; X86-BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; X86-BMI2-NEXT:    testb $32, %cl
+; 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 4(%esi), %edx
+; X86-BMI2-NEXT:    andl (%esi), %eax
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_ic2_load:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $64, %ecx
+; X64-NOBMI2-NEXT:    subl %esi, %ecx
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq (%rdi), %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_ic2_load:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $64, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movq $-1, %rcx
+; X64-BMI2-NEXT:    shlxq %rax, %rcx, %rax
+; X64-BMI2-NEXT:    andq (%rdi), %rax
+; X64-BMI2-NEXT:    retq
+  %val = load i64, i64* %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(i64* %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:    shldl %cl, %edx, %edx
+; 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 4(%esi), %edx
+; X86-NOBMI2-NEXT:    andl (%esi), %eax
+; 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 %esi
+; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; 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:    shldl %cl, %edx, %edx
+; X86-BMI2-NEXT:    testb $32, %cl
+; 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 4(%esi), %edx
+; X86-BMI2-NEXT:    andl (%esi), %eax
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movb $64, %cl
+; X64-NOBMI2-NEXT:    subb %sil, %cl
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq (%rdi), %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movb $64, %al
+; X64-BMI2-NEXT:    subb %sil, %al
+; X64-BMI2-NEXT:    movq $-1, %rcx
+; X64-BMI2-NEXT:    shlxq %rax, %rcx, %rax
+; X64-BMI2-NEXT:    andq (%rdi), %rax
+; X64-BMI2-NEXT:    retq
+  %val = load i64, i64* %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:    movl $64, %ecx
+; X86-NOBMI2-NEXT:    subl {{[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:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: clear_lowbits64_ic4_commutative:
+; X86-BMI2:       # %bb.0:
+; X86-BMI2-NEXT:    movl $64, %ecx
+; X86-BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT:    movl $-1, %edx
+; X86-BMI2-NEXT:    shlxl %ecx, %edx, %eax
+; X86-BMI2-NEXT:    shldl %cl, %edx, %edx
+; 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), %edx
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: clear_lowbits64_ic4_commutative:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    movl $64, %ecx
+; X64-NOBMI2-NEXT:    subl %esi, %ecx
+; X64-NOBMI2-NEXT:    movq $-1, %rax
+; X64-NOBMI2-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rax
+; X64-NOBMI2-NEXT:    andq %rdi, %rax
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: clear_lowbits64_ic4_commutative:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    movl $64, %eax
+; X64-BMI2-NEXT:    subl %esi, %eax
+; X64-BMI2-NEXT:    movq $-1, %rcx
+; X64-BMI2-NEXT:    shlxq %rax, %rcx, %rax
+; X64-BMI2-NEXT:    andq %rdi, %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
+; ---------------------------------------------------------------------------- ;
+
+declare void @use32(i32)
+declare void @use64(i64)
+
+define i32 @oneuse32(i32 %val, i32 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: oneuse32:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    subl $8, %esp
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %esi
+; X86-NOBMI2-NEXT:    shll %cl, %esi
+; X86-NOBMI2-NEXT:    movl %esi, (%esp)
+; X86-NOBMI2-NEXT:    calll use32
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    movl %esi, %eax
+; X86-NOBMI2-NEXT:    addl $8, %esp
+; X86-NOBMI2-NEXT:    popl %esi
+; X86-NOBMI2-NEXT:    retl
+;
+; X86-BMI2-LABEL: oneuse32:
+; 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 $-1, %ecx
+; X86-BMI2-NEXT:    shlxl %eax, %ecx, %esi
+; X86-BMI2-NEXT:    movl %esi, (%esp)
+; X86-BMI2-NEXT:    calll use32
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    movl %esi, %eax
+; X86-BMI2-NEXT:    addl $8, %esp
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: oneuse32:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    pushq %rbp
+; X64-NOBMI2-NEXT:    pushq %rbx
+; X64-NOBMI2-NEXT:    pushq %rax
+; X64-NOBMI2-NEXT:    movl %edi, %ebx
+; X64-NOBMI2-NEXT:    movl $-1, %ebp
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shll %cl, %ebp
+; X64-NOBMI2-NEXT:    movl %ebp, %edi
+; X64-NOBMI2-NEXT:    callq use32
+; X64-NOBMI2-NEXT:    andl %ebx, %ebp
+; X64-NOBMI2-NEXT:    movl %ebp, %eax
+; X64-NOBMI2-NEXT:    addq $8, %rsp
+; X64-NOBMI2-NEXT:    popq %rbx
+; X64-NOBMI2-NEXT:    popq %rbp
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: oneuse32:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    pushq %rbp
+; X64-BMI2-NEXT:    pushq %rbx
+; X64-BMI2-NEXT:    pushq %rax
+; X64-BMI2-NEXT:    movl %edi, %ebx
+; X64-BMI2-NEXT:    movl $-1, %eax
+; X64-BMI2-NEXT:    shlxl %esi, %eax, %ebp
+; X64-BMI2-NEXT:    movl %ebp, %edi
+; X64-BMI2-NEXT:    callq use32
+; X64-BMI2-NEXT:    andl %ebx, %ebp
+; X64-BMI2-NEXT:    movl %ebp, %eax
+; X64-BMI2-NEXT:    addq $8, %rsp
+; X64-BMI2-NEXT:    popq %rbx
+; X64-BMI2-NEXT:    popq %rbp
+; X64-BMI2-NEXT:    retq
+  %mask = shl i32 -1, %numlowbits
+  call void @use32(i32 %mask)
+  %masked = and i32 %mask, %val
+  ret i32 %masked
+}
+
+define i64 @oneuse64(i64 %val, i64 %numlowbits) nounwind {
+; X86-NOBMI2-LABEL: oneuse64:
+; X86-NOBMI2:       # %bb.0:
+; X86-NOBMI2-NEXT:    pushl %edi
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    pushl %eax
+; X86-NOBMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI2-NEXT:    movl $-1, %esi
+; X86-NOBMI2-NEXT:    movl $-1, %edi
+; X86-NOBMI2-NEXT:    shll %cl, %edi
+; X86-NOBMI2-NEXT:    shldl %cl, %esi, %esi
+; X86-NOBMI2-NEXT:    testb $32, %cl
+; X86-NOBMI2-NEXT:    je .LBB37_2
+; X86-NOBMI2-NEXT:  # %bb.1:
+; X86-NOBMI2-NEXT:    movl %edi, %esi
+; X86-NOBMI2-NEXT:    xorl %edi, %edi
+; X86-NOBMI2-NEXT:  .LBB37_2:
+; X86-NOBMI2-NEXT:    subl $8, %esp
+; X86-NOBMI2-NEXT:    pushl %esi
+; X86-NOBMI2-NEXT:    pushl %edi
+; X86-NOBMI2-NEXT:    calll use64
+; X86-NOBMI2-NEXT:    addl $16, %esp
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI2-NEXT:    andl {{[0-9]+}}(%esp), %edi
+; X86-NOBMI2-NEXT:    movl %edi, %eax
+; X86-NOBMI2-NEXT:    movl %esi, %edx
+; X86-NOBMI2-NEXT:    addl $4, %esp
+; 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 %edi
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    pushl %eax
+; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT:    movl $-1, %esi
+; X86-BMI2-NEXT:    shlxl %ecx, %esi, %edi
+; X86-BMI2-NEXT:    shldl %cl, %esi, %esi
+; X86-BMI2-NEXT:    testb $32, %cl
+; X86-BMI2-NEXT:    je .LBB37_2
+; X86-BMI2-NEXT:  # %bb.1:
+; X86-BMI2-NEXT:    movl %edi, %esi
+; X86-BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI2-NEXT:  .LBB37_2:
+; X86-BMI2-NEXT:    subl $8, %esp
+; X86-BMI2-NEXT:    pushl %esi
+; X86-BMI2-NEXT:    pushl %edi
+; X86-BMI2-NEXT:    calll use64
+; X86-BMI2-NEXT:    addl $16, %esp
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
+; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %edi
+; X86-BMI2-NEXT:    movl %edi, %eax
+; X86-BMI2-NEXT:    movl %esi, %edx
+; X86-BMI2-NEXT:    addl $4, %esp
+; X86-BMI2-NEXT:    popl %esi
+; X86-BMI2-NEXT:    popl %edi
+; X86-BMI2-NEXT:    retl
+;
+; X64-NOBMI2-LABEL: oneuse64:
+; X64-NOBMI2:       # %bb.0:
+; X64-NOBMI2-NEXT:    pushq %r14
+; X64-NOBMI2-NEXT:    pushq %rbx
+; X64-NOBMI2-NEXT:    pushq %rax
+; X64-NOBMI2-NEXT:    movq %rdi, %r14
+; X64-NOBMI2-NEXT:    movq $-1, %rbx
+; X64-NOBMI2-NEXT:    movl %esi, %ecx
+; X64-NOBMI2-NEXT:    shlq %cl, %rbx
+; X64-NOBMI2-NEXT:    movq %rbx, %rdi
+; X64-NOBMI2-NEXT:    callq use64
+; X64-NOBMI2-NEXT:    andq %r14, %rbx
+; X64-NOBMI2-NEXT:    movq %rbx, %rax
+; X64-NOBMI2-NEXT:    addq $8, %rsp
+; X64-NOBMI2-NEXT:    popq %rbx
+; X64-NOBMI2-NEXT:    popq %r14
+; X64-NOBMI2-NEXT:    retq
+;
+; X64-BMI2-LABEL: oneuse64:
+; X64-BMI2:       # %bb.0:
+; X64-BMI2-NEXT:    pushq %r14
+; X64-BMI2-NEXT:    pushq %rbx
+; X64-BMI2-NEXT:    pushq %rax
+; X64-BMI2-NEXT:    movq %rdi, %r14
+; X64-BMI2-NEXT:    movq $-1, %rax
+; X64-BMI2-NEXT:    shlxq %rsi, %rax, %rbx
+; X64-BMI2-NEXT:    movq %rbx, %rdi
+; X64-BMI2-NEXT:    callq use64
+; X64-BMI2-NEXT:    andq %r14, %rbx
+; X64-BMI2-NEXT:    movq %rbx, %rax
+; X64-BMI2-NEXT:    addq $8, %rsp
+; X64-BMI2-NEXT:    popq %rbx
+; X64-BMI2-NEXT:    popq %r14
+; X64-BMI2-NEXT:    retq
+  %mask = shl i64 -1, %numlowbits
+  call void @use64(i64 %mask)
+  %masked = and i64 %mask, %val
+  ret i64 %masked
+}