llvm/test/CodeGen/X86/avx-shift.ll
Bruno Cardoso Lopes 0dd80b0d69 Fix PR10688. Add support for spliting 256-bit vector shifts when the
shift amount is variable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@137885 91177308-0d34-0410-b5e6-96231b3b80d8
2011-08-17 22:12:20 +00:00

76 lines
2.1 KiB
LLVM

; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck %s
;;; Shift left
; CHECK: vpslld
; CHECK: vpslld
define <8 x i32> @vshift00(<8 x i32> %a) nounwind readnone {
%s = shl <8 x i32> %a, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32
2>
ret <8 x i32> %s
}
; CHECK: vpsllw
; CHECK: vpsllw
define <16 x i16> @vshift01(<16 x i16> %a) nounwind readnone {
%s = shl <16 x i16> %a, <i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2>
ret <16 x i16> %s
}
; CHECK: vpsllq
; CHECK: vpsllq
define <4 x i64> @vshift02(<4 x i64> %a) nounwind readnone {
%s = shl <4 x i64> %a, <i64 2, i64 2, i64 2, i64 2>
ret <4 x i64> %s
}
;;; Logical Shift right
; CHECK: vpsrld
; CHECK: vpsrld
define <8 x i32> @vshift03(<8 x i32> %a) nounwind readnone {
%s = lshr <8 x i32> %a, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32
2>
ret <8 x i32> %s
}
; CHECK: vpsrlw
; CHECK: vpsrlw
define <16 x i16> @vshift04(<16 x i16> %a) nounwind readnone {
%s = lshr <16 x i16> %a, <i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2>
ret <16 x i16> %s
}
; CHECK: vpsrlq
; CHECK: vpsrlq
define <4 x i64> @vshift05(<4 x i64> %a) nounwind readnone {
%s = lshr <4 x i64> %a, <i64 2, i64 2, i64 2, i64 2>
ret <4 x i64> %s
}
;;; Arithmetic Shift right
; CHECK: vpsrad
; CHECK: vpsrad
define <8 x i32> @vshift06(<8 x i32> %a) nounwind readnone {
%s = ashr <8 x i32> %a, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32
2>
ret <8 x i32> %s
}
; CHECK: vpsraw
; CHECK: vpsraw
define <16 x i16> @vshift07(<16 x i16> %a) nounwind readnone {
%s = ashr <16 x i16> %a, <i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2>
ret <16 x i16> %s
}
;;; Support variable shifts
; CHECK: _vshift08
; CHECK: vextractf128 $1
; CHECK: vpslld $23
; CHECK: vextractf128 $1
; CHECK: vpslld $23
define <8 x i32> @vshift08(<8 x i32> %a) nounwind {
%bitop = shl <8 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>, %a
ret <8 x i32> %bitop
}