llvm/test/CodeGen/ARM/2010-05-21-BuildVector.ll
Bob Wilson be751cfe9c Recognize more BUILD_VECTORs and VECTOR_SHUFFLEs that can be implemented by
copying VFP subregs.  This exposed a bunch of dead code in the *spill-q.ll
tests, so I tweaked those tests to keep that code from being optimized away.
Radar 7872877.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104415 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-22 00:23:12 +00:00

44 lines
1.5 KiB
LLVM

; RUN: llc < %s -march=arm -mcpu=cortex-a8 | FileCheck %s
; Radar 7872877
define arm_apcscc void @test(float* %fltp, i32 %packedValue, float* %table) nounwind {
entry:
%0 = load float* %fltp
%1 = insertelement <4 x float> undef, float %0, i32 0
%2 = shufflevector <4 x float> %1, <4 x float> undef, <4 x i32> zeroinitializer
%3 = shl i32 %packedValue, 16
%4 = ashr i32 %3, 30
%.sum = add i32 %4, 4
%5 = getelementptr inbounds float* %table, i32 %.sum
;CHECK: vldr.32 s
%6 = load float* %5, align 4
%tmp11 = insertelement <4 x float> undef, float %6, i32 0
%7 = shl i32 %packedValue, 18
%8 = ashr i32 %7, 30
%.sum12 = add i32 %8, 4
%9 = getelementptr inbounds float* %table, i32 %.sum12
;CHECK: vldr.32 s
%10 = load float* %9, align 4
%tmp9 = insertelement <4 x float> %tmp11, float %10, i32 1
%11 = shl i32 %packedValue, 20
%12 = ashr i32 %11, 30
%.sum13 = add i32 %12, 4
%13 = getelementptr inbounds float* %table, i32 %.sum13
;CHECK: vldr.32 s
%14 = load float* %13, align 4
%tmp7 = insertelement <4 x float> %tmp9, float %14, i32 2
%15 = shl i32 %packedValue, 22
%16 = ashr i32 %15, 30
%.sum14 = add i32 %16, 4
%17 = getelementptr inbounds float* %table, i32 %.sum14
;CHECK: vldr.32 s
%18 = load float* %17, align 4
%tmp5 = insertelement <4 x float> %tmp7, float %18, i32 3
%19 = fmul <4 x float> %tmp5, %2
%20 = bitcast float* %fltp to i8*
tail call void @llvm.arm.neon.vst1.v4f32(i8* %20, <4 x float> %19)
ret void
}
declare void @llvm.arm.neon.vst1.v4f32(i8*, <4 x float>) nounwind