[llvm] fbf9356 - [X86] Add ctpop-mask.ll - test coverage based off #79823
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Fri Feb 2 07:40:39 PST 2024
Author: Simon Pilgrim
Date: 2024-02-02T15:40:22Z
New Revision: fbf9356be07408efc05621a7a8d0472a61c66228
URL: https://github.com/llvm/llvm-project/commit/fbf9356be07408efc05621a7a8d0472a61c66228
DIFF: https://github.com/llvm/llvm-project/commit/fbf9356be07408efc05621a7a8d0472a61c66228.diff
LOG: [X86] Add ctpop-mask.ll - test coverage based off #79823
Added:
llvm/test/CodeGen/X86/ctpop-mask.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/X86/ctpop-mask.ll b/llvm/test/CodeGen/X86/ctpop-mask.ll
new file mode 100644
index 0000000000000..1af964dd11c66
--- /dev/null
+++ b/llvm/test/CodeGen/X86/ctpop-mask.ll
@@ -0,0 +1,824 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown -mattr=+popcnt | FileCheck %s -check-prefixes=X86-POPCOUNT
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+popcnt | FileCheck %s -check-prefixes=X64-POPCOUNT
+; RUN: llc < %s -mtriple=i686-unknown -mattr=-popcnt | FileCheck %s -check-prefixes=X86-NO-POPCOUNT
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=-popcnt | FileCheck %s -check-prefixes=X64-NO-POPCOUNT
+
+declare i8 @llvm.ctpop.i8(i8) nounwind readnone
+declare i16 @llvm.ctpop.i16(i16) nounwind readnone
+declare i32 @llvm.ctpop.i32(i32) nounwind readnone
+declare i64 @llvm.ctpop.i64(i64) nounwind readnone
+
+; PR79823
+; CTPOP tests where we know only some bits are non-zero.
+; FIXME: Move shifted masked values to lsb to use smaller bitsize special cases.
+
+define i64 @ctpop_mask2(i64 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask2:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: andl $3, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask2:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $3, %edi
+; X64-POPCOUNT-NEXT: popcntq %rdi, %rax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask2:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: andl $3, %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask2:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: andl $3, %edi
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %edi, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i64 %x, 3
+ %count = tail call i64 @llvm.ctpop.i64(i64 %mask)
+ ret i64 %count
+}
+
+define i32 @ctpop_shifted_mask2(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shifted_mask2:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl $1572864, %eax # imm = 0x180000
+; X86-POPCOUNT-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shifted_mask2:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $1572864, %edi # imm = 0x180000
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shifted_mask2:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NO-POPCOUNT-NEXT: movl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: andl $1572864, %eax # imm = 0x180000
+; X86-NO-POPCOUNT-NEXT: shrl %ecx
+; X86-NO-POPCOUNT-NEXT: andl $262144, %ecx # imm = 0x40000
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858783744, %ecx # imm = 0x33300000
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858980352, %eax # imm = 0x33330000
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shifted_mask2:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: andl $1572864, %eax # imm = 0x180000
+; X64-NO-POPCOUNT-NEXT: shrl %edi
+; X64-NO-POPCOUNT-NEXT: andl $262144, %edi # imm = 0x40000
+; X64-NO-POPCOUNT-NEXT: subl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $858783744, %ecx # imm = 0x33300000
+; X64-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X64-NO-POPCOUNT-NEXT: andl $858980352, %eax # imm = 0x33330000
+; X64-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X64-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X64-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X64-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 1572864 ; 3 << 19
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i32 @ctpop_mask3(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask3:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: andl $5, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask3:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $5, %edi
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask3:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: andl $5, %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask3:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: andl $5, %edi
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %edi, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 5 ; 0b101
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i16 @ctpop_shifted_mask3(i16 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shifted_mask3:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: andl $14, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shifted_mask3:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $14, %edi
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shifted_mask3:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: andl $14, %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shifted_mask3:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: andl $14, %edi
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %edi, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i16 %x, 14 ; 7 << 1
+ %count = tail call i16 @llvm.ctpop.i16(i16 %mask)
+ ret i16 %count
+}
+
+define i64 @ctpop_mask4(i64 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask4:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: andl $15, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask4:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $15, %edi
+; X64-POPCOUNT-NEXT: popcntq %rdi, %rax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask4:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: andl $15, %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask4:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: andl $15, %edi
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %edi, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $17895697, %eax # imm = 0x1111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i64 %x, 15
+ %count = tail call i64 @llvm.ctpop.i64(i64 %mask)
+ ret i64 %count
+}
+
+define i32 @ctpop_shifted_mask4(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shifted_mask4:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl $7680, %eax # imm = 0x1E00
+; X86-POPCOUNT-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shifted_mask4:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $7680, %edi # imm = 0x1E00
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shifted_mask4:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NO-POPCOUNT-NEXT: movl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: andl $7680, %eax # imm = 0x1E00
+; X86-NO-POPCOUNT-NEXT: shrl %ecx
+; X86-NO-POPCOUNT-NEXT: andl $1280, %ecx # imm = 0x500
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %ecx # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %eax # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shifted_mask4:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: andl $7680, %eax # imm = 0x1E00
+; X64-NO-POPCOUNT-NEXT: shrl %edi
+; X64-NO-POPCOUNT-NEXT: andl $1280, %edi # imm = 0x500
+; X64-NO-POPCOUNT-NEXT: subl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $858993408, %ecx # imm = 0x33333300
+; X64-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X64-NO-POPCOUNT-NEXT: andl $858993408, %eax # imm = 0x33333300
+; X64-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X64-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X64-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X64-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 7680 ; 15 << 9
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i64 @ctpop_mask5(i64 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask5:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: andl $31, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask5:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $31, %edi
+; X64-POPCOUNT-NEXT: popcntq %rdi, %rax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask5:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: andl $31, %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask5:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: andl $31, %edi
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %edi, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i64 %x, 31
+ %count = tail call i64 @llvm.ctpop.i64(i64 %mask)
+ ret i64 %count
+}
+
+define i32 @ctpop_shifted_mask5(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shifted_mask5:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl $11776, %eax # imm = 0x2E00
+; X86-POPCOUNT-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shifted_mask5:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $11776, %edi # imm = 0x2E00
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shifted_mask5:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NO-POPCOUNT-NEXT: movl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: andl $11776, %eax # imm = 0x2E00
+; X86-NO-POPCOUNT-NEXT: shrl %ecx
+; X86-NO-POPCOUNT-NEXT: andl $5376, %ecx # imm = 0x1500
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %ecx # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %eax # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shifted_mask5:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: andl $11776, %eax # imm = 0x2E00
+; X64-NO-POPCOUNT-NEXT: shrl %edi
+; X64-NO-POPCOUNT-NEXT: andl $5376, %edi # imm = 0x1500
+; X64-NO-POPCOUNT-NEXT: subl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $858993408, %ecx # imm = 0x33333300
+; X64-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X64-NO-POPCOUNT-NEXT: andl $858993408, %eax # imm = 0x33333300
+; X64-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X64-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X64-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X64-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 11776 ; 23 << 9
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i32 @ctpop_mask6(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask6:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: andl $63, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask6:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $63, %edi
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask6:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: andl $63, %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask6:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: andl $63, %edi
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %edi, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 63
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i64 @ctpop_shifted_mask6(i64 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shifted_mask6:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl $26112, %eax # imm = 0x6600
+; X86-POPCOUNT-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shifted_mask6:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $26112, %edi # imm = 0x6600
+; X64-POPCOUNT-NEXT: popcntq %rdi, %rax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shifted_mask6:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NO-POPCOUNT-NEXT: movl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: andl $26112, %eax # imm = 0x6600
+; X86-NO-POPCOUNT-NEXT: shrl %ecx
+; X86-NO-POPCOUNT-NEXT: andl $4352, %ecx # imm = 0x1100
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %ecx # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %eax # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shifted_mask6:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: andl $26112, %eax # imm = 0x6600
+; X64-NO-POPCOUNT-NEXT: shrl %edi
+; X64-NO-POPCOUNT-NEXT: andl $4352, %edi # imm = 0x1100
+; X64-NO-POPCOUNT-NEXT: subq %rdi, %rax
+; X64-NO-POPCOUNT-NEXT: movabsq $3689348814741910272, %rcx # imm = 0x3333333333333300
+; X64-NO-POPCOUNT-NEXT: movq %rax, %rdx
+; X64-NO-POPCOUNT-NEXT: andq %rcx, %rdx
+; X64-NO-POPCOUNT-NEXT: shrq $2, %rax
+; X64-NO-POPCOUNT-NEXT: andq %rcx, %rax
+; X64-NO-POPCOUNT-NEXT: addq %rdx, %rax
+; X64-NO-POPCOUNT-NEXT: movq %rax, %rcx
+; X64-NO-POPCOUNT-NEXT: shrq $4, %rcx
+; X64-NO-POPCOUNT-NEXT: addq %rax, %rcx
+; X64-NO-POPCOUNT-NEXT: movabsq $1085102592571150095, %rdx # imm = 0xF0F0F0F0F0F0F0F
+; X64-NO-POPCOUNT-NEXT: andq %rcx, %rdx
+; X64-NO-POPCOUNT-NEXT: movabsq $72340172838076673, %rax # imm = 0x101010101010101
+; X64-NO-POPCOUNT-NEXT: imulq %rdx, %rax
+; X64-NO-POPCOUNT-NEXT: shrq $56, %rax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i64 %x, 26112 ; 51 << 9
+ %count = tail call i64 @llvm.ctpop.i64(i64 %mask)
+ ret i64 %count
+}
+
+define i16 @ctpop_mask7(i16 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask7:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: andl $127, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask7:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $127, %edi
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask7:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: andl $127, %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask7:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: andl $127, %edi
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %edi, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: # kill: def $ax killed $ax killed $eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i16 %x, 127
+ %count = tail call i16 @llvm.ctpop.i16(i16 %mask)
+ ret i16 %count
+}
+
+define i32 @ctpop_shift_mask7(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shift_mask7:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl $1040384, %eax # imm = 0xFE000
+; X86-POPCOUNT-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shift_mask7:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $1040384, %edi # imm = 0xFE000
+; X64-POPCOUNT-NEXT: popcntl %edi, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shift_mask7:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NO-POPCOUNT-NEXT: movl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: andl $1040384, %eax # imm = 0xFE000
+; X86-NO-POPCOUNT-NEXT: shrl %ecx
+; X86-NO-POPCOUNT-NEXT: andl $348160, %ecx # imm = 0x55000
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858992640, %ecx # imm = 0x33333000
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858992640, %eax # imm = 0x33333000
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shift_mask7:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: andl $1040384, %eax # imm = 0xFE000
+; X64-NO-POPCOUNT-NEXT: shrl %edi
+; X64-NO-POPCOUNT-NEXT: andl $348160, %edi # imm = 0x55000
+; X64-NO-POPCOUNT-NEXT: subl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $858992640, %ecx # imm = 0x33333000
+; X64-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X64-NO-POPCOUNT-NEXT: andl $858992640, %eax # imm = 0x33333000
+; X64-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X64-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X64-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X64-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 1040384 ; 127 << 13
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i32 @ctpop_mask8(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask8:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movzbl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask8:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: movzbl %dil, %eax
+; X64-POPCOUNT-NEXT: popcntl %eax, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask8:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movzbl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X86-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X86-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X86-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask8:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movzbl %dil, %eax
+; X64-NO-POPCOUNT-NEXT: imull $134480385, %eax, %eax # imm = 0x8040201
+; X64-NO-POPCOUNT-NEXT: shrl $3, %eax
+; X64-NO-POPCOUNT-NEXT: andl $286331153, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: imull $286331153, %eax, %eax # imm = 0x11111111
+; X64-NO-POPCOUNT-NEXT: shrl $28, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 255
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i64 @ctpop_shifted_mask8(i64 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shifted_mask8:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movzbl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: shll $8, %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shifted_mask8:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: andl $65280, %edi # imm = 0xFF00
+; X64-POPCOUNT-NEXT: popcntq %rdi, %rax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shifted_mask8:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
+; X86-NO-POPCOUNT-NEXT: movl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: shll $8, %eax
+; X86-NO-POPCOUNT-NEXT: shll $7, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $21760, %ecx # imm = 0x5500
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %ecx # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858993408, %eax # imm = 0x33333300
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shifted_mask8:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: andl $65280, %eax # imm = 0xFF00
+; X64-NO-POPCOUNT-NEXT: shrl %edi
+; X64-NO-POPCOUNT-NEXT: andl $21760, %edi # imm = 0x5500
+; X64-NO-POPCOUNT-NEXT: subq %rdi, %rax
+; X64-NO-POPCOUNT-NEXT: movabsq $3689348814741910272, %rcx # imm = 0x3333333333333300
+; X64-NO-POPCOUNT-NEXT: movq %rax, %rdx
+; X64-NO-POPCOUNT-NEXT: andq %rcx, %rdx
+; X64-NO-POPCOUNT-NEXT: shrq $2, %rax
+; X64-NO-POPCOUNT-NEXT: andq %rcx, %rax
+; X64-NO-POPCOUNT-NEXT: addq %rdx, %rax
+; X64-NO-POPCOUNT-NEXT: movq %rax, %rcx
+; X64-NO-POPCOUNT-NEXT: shrq $4, %rcx
+; X64-NO-POPCOUNT-NEXT: addq %rax, %rcx
+; X64-NO-POPCOUNT-NEXT: movabsq $1085102592571150095, %rdx # imm = 0xF0F0F0F0F0F0F0F
+; X64-NO-POPCOUNT-NEXT: andq %rcx, %rdx
+; X64-NO-POPCOUNT-NEXT: movabsq $72340172838076673, %rax # imm = 0x101010101010101
+; X64-NO-POPCOUNT-NEXT: imulq %rdx, %rax
+; X64-NO-POPCOUNT-NEXT: shrq $56, %rax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i64 %x, 65280 ; 255 << 8
+ %count = tail call i64 @llvm.ctpop.i64(i64 %mask)
+ ret i64 %count
+}
+
+define i32 @ctpop_mask16(i32 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_mask16:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_mask16:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: movzwl %di, %eax
+; X64-POPCOUNT-NEXT: popcntl %eax, %eax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_mask16:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl %ecx
+; X86-NO-POPCOUNT-NEXT: andl $21845, %ecx # imm = 0x5555
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858993459, %ecx # imm = 0x33333333
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858993459, %eax # imm = 0x33333333
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_mask16:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movzwl %di, %eax
+; X64-NO-POPCOUNT-NEXT: shrl %edi
+; X64-NO-POPCOUNT-NEXT: andl $21845, %edi # imm = 0x5555
+; X64-NO-POPCOUNT-NEXT: subl %edi, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $858993459, %ecx # imm = 0x33333333
+; X64-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X64-NO-POPCOUNT-NEXT: andl $858993459, %eax # imm = 0x33333333
+; X64-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X64-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X64-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X64-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X64-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X64-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i32 %x, 65535
+ %count = tail call i32 @llvm.ctpop.i32(i32 %mask)
+ ret i32 %count
+}
+
+define i64 @ctpop_shifted_mask16(i64 %x) nounwind readnone {
+; X86-POPCOUNT-LABEL: ctpop_shifted_mask16:
+; X86-POPCOUNT: # %bb.0:
+; X86-POPCOUNT-NEXT: movl $524280, %eax # imm = 0x7FFF8
+; X86-POPCOUNT-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-POPCOUNT-NEXT: popcntl %eax, %eax
+; X86-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-POPCOUNT-NEXT: retl
+;
+; X64-POPCOUNT-LABEL: ctpop_shifted_mask16:
+; X64-POPCOUNT: # %bb.0:
+; X64-POPCOUNT-NEXT: movabsq $2251765453946880, %rax # imm = 0x7FFF800000000
+; X64-POPCOUNT-NEXT: andq %rdi, %rax
+; X64-POPCOUNT-NEXT: popcntq %rax, %rax
+; X64-POPCOUNT-NEXT: retq
+;
+; X86-NO-POPCOUNT-LABEL: ctpop_shifted_mask16:
+; X86-NO-POPCOUNT: # %bb.0:
+; X86-NO-POPCOUNT-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-NO-POPCOUNT-NEXT: movl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: andl $524280, %eax # imm = 0x7FFF8
+; X86-NO-POPCOUNT-NEXT: shrl %ecx
+; X86-NO-POPCOUNT-NEXT: andl $87380, %ecx # imm = 0x15554
+; X86-NO-POPCOUNT-NEXT: subl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $858993456, %ecx # imm = 0x33333330
+; X86-NO-POPCOUNT-NEXT: shrl $2, %eax
+; X86-NO-POPCOUNT-NEXT: andl $858993459, %eax # imm = 0x33333333
+; X86-NO-POPCOUNT-NEXT: addl %ecx, %eax
+; X86-NO-POPCOUNT-NEXT: movl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: shrl $4, %ecx
+; X86-NO-POPCOUNT-NEXT: addl %eax, %ecx
+; X86-NO-POPCOUNT-NEXT: andl $252645135, %ecx # imm = 0xF0F0F0F
+; X86-NO-POPCOUNT-NEXT: imull $16843009, %ecx, %eax # imm = 0x1010101
+; X86-NO-POPCOUNT-NEXT: shrl $24, %eax
+; X86-NO-POPCOUNT-NEXT: xorl %edx, %edx
+; X86-NO-POPCOUNT-NEXT: retl
+;
+; X64-NO-POPCOUNT-LABEL: ctpop_shifted_mask16:
+; X64-NO-POPCOUNT: # %bb.0:
+; X64-NO-POPCOUNT-NEXT: movabsq $2251765453946880, %rax # imm = 0x7FFF800000000
+; X64-NO-POPCOUNT-NEXT: andq %rdi, %rax
+; X64-NO-POPCOUNT-NEXT: shrq %rdi
+; X64-NO-POPCOUNT-NEXT: movabsq $375294242324480, %rcx # imm = 0x1555400000000
+; X64-NO-POPCOUNT-NEXT: andq %rdi, %rcx
+; X64-NO-POPCOUNT-NEXT: subq %rcx, %rax
+; X64-NO-POPCOUNT-NEXT: movabsq $3689348800998014976, %rcx # imm = 0x3333333000000000
+; X64-NO-POPCOUNT-NEXT: andq %rax, %rcx
+; X64-NO-POPCOUNT-NEXT: shrq $2, %rax
+; X64-NO-POPCOUNT-NEXT: movabsq $3689348813882916864, %rdx # imm = 0x3333333300000000
+; X64-NO-POPCOUNT-NEXT: andq %rax, %rdx
+; X64-NO-POPCOUNT-NEXT: addq %rcx, %rdx
+; X64-NO-POPCOUNT-NEXT: movq %rdx, %rax
+; X64-NO-POPCOUNT-NEXT: shrq $4, %rax
+; X64-NO-POPCOUNT-NEXT: addq %rdx, %rax
+; X64-NO-POPCOUNT-NEXT: movabsq $1085102592571150095, %rcx # imm = 0xF0F0F0F0F0F0F0F
+; X64-NO-POPCOUNT-NEXT: andq %rax, %rcx
+; X64-NO-POPCOUNT-NEXT: movabsq $72340172838076673, %rax # imm = 0x101010101010101
+; X64-NO-POPCOUNT-NEXT: imulq %rcx, %rax
+; X64-NO-POPCOUNT-NEXT: shrq $56, %rax
+; X64-NO-POPCOUNT-NEXT: retq
+ %mask = and i64 %x, 2251765453946880 ; 65535 << 35
+ %count = tail call i64 @llvm.ctpop.i64(i64 %mask)
+ ret i64 %count
+}
More information about the llvm-commits
mailing list