8779b11410
We have this generic transform in IR (instcombine), but as shown in PR41098: http://bugs.llvm.org/PR41098 ...the pattern may emerge in codegen too. x86 has a potential refinement/reversal opportunity here, but that should come later or needs a target hook to avoid the transform. Converting to bswap is the more specific form, so we should use it if it is available.
335 lines
9.0 KiB
LLVM
335 lines
9.0 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefixes=X32,BASE32
|
|
; RUN: llc < %s -mtriple=i686-- -mattr=movbe | FileCheck %s --check-prefixes=X32,MOVBE32
|
|
; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefixes=X64,BASE64
|
|
; RUN: llc < %s -mtriple=x86_64-- -mattr=movbe | FileCheck %s --check-prefixes=X64,MOVBE64
|
|
|
|
define i16 @foo(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: foo:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movb {{[0-9]+}}(%esp), %cl
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: rolw %cl, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: foo:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edx, %ecx
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: # kill: def $cl killed $cl killed $ecx
|
|
; X64-NEXT: rolw %cl, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = shl i16 %x, %z
|
|
%t1 = sub i16 16, %z
|
|
%t2 = lshr i16 %x, %t1
|
|
%t3 = or i16 %t2, %t0
|
|
ret i16 %t3
|
|
}
|
|
|
|
define i16 @bar(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: bar:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %edx
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: movb {{[0-9]+}}(%esp), %cl
|
|
; X32-NEXT: andb $15, %cl
|
|
; X32-NEXT: shldw %cl, %dx, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: bar:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edx, %ecx
|
|
; X64-NEXT: movl %esi, %eax
|
|
; X64-NEXT: andb $15, %cl
|
|
; X64-NEXT: # kill: def $cl killed $cl killed $ecx
|
|
; X64-NEXT: shldw %cl, %di, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = shl i16 %y, %z
|
|
%t1 = sub i16 16, %z
|
|
%t2 = lshr i16 %x, %t1
|
|
%t3 = or i16 %t2, %t0
|
|
ret i16 %t3
|
|
}
|
|
|
|
define i16 @un(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: un:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movb {{[0-9]+}}(%esp), %cl
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: rorw %cl, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: un:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edx, %ecx
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: # kill: def $cl killed $cl killed $ecx
|
|
; X64-NEXT: rorw %cl, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = lshr i16 %x, %z
|
|
%t1 = sub i16 16, %z
|
|
%t2 = shl i16 %x, %t1
|
|
%t3 = or i16 %t2, %t0
|
|
ret i16 %t3
|
|
}
|
|
|
|
define i16 @bu(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: bu:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %edx
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: movb {{[0-9]+}}(%esp), %cl
|
|
; X32-NEXT: andb $15, %cl
|
|
; X32-NEXT: shrdw %cl, %dx, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: bu:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edx, %ecx
|
|
; X64-NEXT: movl %esi, %eax
|
|
; X64-NEXT: andb $15, %cl
|
|
; X64-NEXT: # kill: def $cl killed $cl killed $ecx
|
|
; X64-NEXT: shrdw %cl, %di, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = lshr i16 %y, %z
|
|
%t1 = sub i16 16, %z
|
|
%t2 = shl i16 %x, %t1
|
|
%t3 = or i16 %t2, %t0
|
|
ret i16 %t3
|
|
}
|
|
|
|
define i16 @xfoo(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: xfoo:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: rolw $5, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: xfoo:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: rolw $5, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = lshr i16 %x, 11
|
|
%t1 = shl i16 %x, 5
|
|
%t2 = or i16 %t0, %t1
|
|
ret i16 %t2
|
|
}
|
|
|
|
define i16 @xbar(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: xbar:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: shldw $5, %cx, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: xbar:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %esi, %eax
|
|
; X64-NEXT: shldw $5, %di, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = shl i16 %y, 5
|
|
%t1 = lshr i16 %x, 11
|
|
%t2 = or i16 %t0, %t1
|
|
ret i16 %t2
|
|
}
|
|
|
|
define i16 @xun(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: xun:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: rolw $11, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: xun:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: rolw $11, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = lshr i16 %x, 5
|
|
%t1 = shl i16 %x, 11
|
|
%t2 = or i16 %t0, %t1
|
|
ret i16 %t2
|
|
}
|
|
|
|
define i16 @xbu(i16 %x, i16 %y, i16 %z) nounwind {
|
|
; X32-LABEL: xbu:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
|
|
; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: shldw $11, %cx, %ax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: xbu:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: shldw $11, %si, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = lshr i16 %y, 5
|
|
%t1 = shl i16 %x, 11
|
|
%t2 = or i16 %t0, %t1
|
|
ret i16 %t2
|
|
}
|
|
|
|
define i32 @rot16_demandedbits(i32 %x, i32 %y) nounwind {
|
|
; X32-LABEL: rot16_demandedbits:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: movl %eax, %ecx
|
|
; X32-NEXT: shrl $11, %ecx
|
|
; X32-NEXT: shll $5, %eax
|
|
; X32-NEXT: orl %ecx, %eax
|
|
; X32-NEXT: movzwl %ax, %eax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: rot16_demandedbits:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: shrl $11, %eax
|
|
; X64-NEXT: shll $5, %edi
|
|
; X64-NEXT: orl %eax, %edi
|
|
; X64-NEXT: movzwl %di, %eax
|
|
; X64-NEXT: retq
|
|
%t0 = lshr i32 %x, 11
|
|
%t1 = shl i32 %x, 5
|
|
%t2 = or i32 %t0, %t1
|
|
%t3 = and i32 %t2, 65535
|
|
ret i32 %t3
|
|
}
|
|
|
|
define i16 @rot16_trunc(i32 %x, i32 %y) nounwind {
|
|
; X32-LABEL: rot16_trunc:
|
|
; X32: # %bb.0:
|
|
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; X32-NEXT: movl %eax, %ecx
|
|
; X32-NEXT: shrl $11, %ecx
|
|
; X32-NEXT: shll $5, %eax
|
|
; X32-NEXT: orl %ecx, %eax
|
|
; X32-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X32-NEXT: retl
|
|
;
|
|
; X64-LABEL: rot16_trunc:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: movl %edi, %ecx
|
|
; X64-NEXT: shrl $11, %ecx
|
|
; X64-NEXT: shll $5, %eax
|
|
; X64-NEXT: orl %ecx, %eax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%t0 = lshr i32 %x, 11
|
|
%t1 = shl i32 %x, 5
|
|
%t2 = or i32 %t0, %t1
|
|
%t3 = trunc i32 %t2 to i16
|
|
ret i16 %t3
|
|
}
|
|
|
|
define i16 @rotate16(i16 %x) {
|
|
; BASE32-LABEL: rotate16:
|
|
; BASE32: # %bb.0:
|
|
; BASE32-NEXT: movzwl {{[0-9]+}}(%esp), %eax
|
|
; BASE32-NEXT: rolw $8, %ax
|
|
; BASE32-NEXT: retl
|
|
;
|
|
; MOVBE32-LABEL: rotate16:
|
|
; MOVBE32: # %bb.0:
|
|
; MOVBE32-NEXT: movbew {{[0-9]+}}(%esp), %ax
|
|
; MOVBE32-NEXT: retl
|
|
;
|
|
; X64-LABEL: rotate16:
|
|
; X64: # %bb.0:
|
|
; X64-NEXT: movl %edi, %eax
|
|
; X64-NEXT: rolw $8, %ax
|
|
; X64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; X64-NEXT: retq
|
|
%r = call i16 @llvm.fshl.i16(i16 %x, i16 %x, i16 8)
|
|
ret i16 %r
|
|
}
|
|
|
|
; TODO: Should this always be rolw with memory operand?
|
|
|
|
define void @rotate16_in_place_memory(i8* %p) {
|
|
; BASE32-LABEL: rotate16_in_place_memory:
|
|
; BASE32: # %bb.0:
|
|
; BASE32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; BASE32-NEXT: rolw $8, (%eax)
|
|
; BASE32-NEXT: retl
|
|
;
|
|
; MOVBE32-LABEL: rotate16_in_place_memory:
|
|
; MOVBE32: # %bb.0:
|
|
; MOVBE32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; MOVBE32-NEXT: movzwl (%eax), %ecx
|
|
; MOVBE32-NEXT: movbew %cx, (%eax)
|
|
; MOVBE32-NEXT: retl
|
|
;
|
|
; BASE64-LABEL: rotate16_in_place_memory:
|
|
; BASE64: # %bb.0:
|
|
; BASE64-NEXT: rolw $8, (%rdi)
|
|
; BASE64-NEXT: retq
|
|
;
|
|
; MOVBE64-LABEL: rotate16_in_place_memory:
|
|
; MOVBE64: # %bb.0:
|
|
; MOVBE64-NEXT: movzwl (%rdi), %eax
|
|
; MOVBE64-NEXT: movbew %ax, (%rdi)
|
|
; MOVBE64-NEXT: retq
|
|
%p0 = getelementptr i8, i8* %p, i64 0
|
|
%p1 = getelementptr i8, i8* %p, i64 1
|
|
%i0 = load i8, i8* %p0, align 1
|
|
%i1 = load i8, i8* %p1, align 1
|
|
store i8 %i1, i8* %p0, align 1
|
|
store i8 %i0, i8* %p1, align 1
|
|
ret void
|
|
}
|
|
|
|
define void @rotate16_memory(i8* %p, i8* %q) {
|
|
; BASE32-LABEL: rotate16_memory:
|
|
; BASE32: # %bb.0:
|
|
; BASE32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; BASE32-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; BASE32-NEXT: movzwl (%ecx), %ecx
|
|
; BASE32-NEXT: rolw $8, %cx
|
|
; BASE32-NEXT: movw %cx, (%eax)
|
|
; BASE32-NEXT: retl
|
|
;
|
|
; MOVBE32-LABEL: rotate16_memory:
|
|
; MOVBE32: # %bb.0:
|
|
; MOVBE32-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
; MOVBE32-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; MOVBE32-NEXT: movzwl (%ecx), %ecx
|
|
; MOVBE32-NEXT: movbew %cx, (%eax)
|
|
; MOVBE32-NEXT: retl
|
|
;
|
|
; BASE64-LABEL: rotate16_memory:
|
|
; BASE64: # %bb.0:
|
|
; BASE64-NEXT: movzwl (%rdi), %eax
|
|
; BASE64-NEXT: rolw $8, %ax
|
|
; BASE64-NEXT: movw %ax, (%rsi)
|
|
; BASE64-NEXT: retq
|
|
;
|
|
; MOVBE64-LABEL: rotate16_memory:
|
|
; MOVBE64: # %bb.0:
|
|
; MOVBE64-NEXT: movzwl (%rdi), %eax
|
|
; MOVBE64-NEXT: movbew %ax, (%rsi)
|
|
; MOVBE64-NEXT: retq
|
|
%p0 = getelementptr i8, i8* %p, i64 0
|
|
%p1 = getelementptr i8, i8* %p, i64 1
|
|
%q0 = getelementptr i8, i8* %q, i64 0
|
|
%q1 = getelementptr i8, i8* %q, i64 1
|
|
%i0 = load i8, i8* %p0, align 1
|
|
%i1 = load i8, i8* %p1, align 1
|
|
store i8 %i1, i8* %q0, align 1
|
|
store i8 %i0, i8* %q1, align 1
|
|
ret void
|
|
}
|
|
|
|
declare i16 @llvm.fshl.i16(i16, i16, i16)
|