%"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
%0 = load <4 x i16>* %arg0_uint16x4_t, align 8 ; <<4 x i16>> [#uses=1]
%1 = extractelement <4 x i16> %0, i32 1 ; <i16> [#uses=1]
- store i16 %1, i16* %out_uint16_t, align 2
+ %2 = add i16 %1, %1
+ store i16 %2, i16* %out_uint16_t, align 2
br label %return
return: ; preds = %entry
%"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
%0 = load <8 x i8>* %arg0_uint8x8_t, align 8 ; <<8 x i8>> [#uses=1]
%1 = extractelement <8 x i8> %0, i32 1 ; <i8> [#uses=1]
- store i8 %1, i8* %out_uint8_t, align 1
+ %2 = add i8 %1, %1
+ store i8 %2, i8* %out_uint8_t, align 1
br label %return
return: ; preds = %entry
%"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
%0 = load <8 x i16>* %arg0_uint16x8_t, align 16 ; <<8 x i16>> [#uses=1]
%1 = extractelement <8 x i16> %0, i32 1 ; <i16> [#uses=1]
- store i16 %1, i16* %out_uint16_t, align 2
+ %2 = add i16 %1, %1
+ store i16 %2, i16* %out_uint16_t, align 2
br label %return
return: ; preds = %entry
%"alloca point" = bitcast i32 0 to i32 ; <i32> [#uses=0]
%0 = load <16 x i8>* %arg0_uint8x16_t, align 16 ; <<16 x i8>> [#uses=1]
%1 = extractelement <16 x i8> %0, i32 1 ; <i8> [#uses=1]
- store i8 %1, i8* %out_uint8_t, align 1
+ %2 = add i8 %1, %1
+ store i8 %2, i8* %out_uint8_t, align 1
br label %return
return: ; preds = %entry
define <4 x i32> @vsetQ_lane32(<4 x i32>* %A, i32 %B) nounwind {
;CHECK: vsetQ_lane32:
-;CHECK: vmov.32
+;CHECK: vmov s
%tmp1 = load <4 x i32>* %A
%tmp2 = insertelement <4 x i32> %tmp1, i32 %B, i32 1
ret <4 x i32> %tmp2
%0 = insertelement <2 x float> %arg1_float32x2_t, float %arg0_float32_t, i32 1 ; <<2 x float>> [#uses=1]
ret <2 x float> %0
}
+
+; The llvm extractelement instruction does not require that the lane number
+; be an immediate constant. Make sure a variable lane number is handled.
+
+define i32 @vget_variable_lanes8(<8 x i8>* %A, i32 %B) nounwind {
+ %tmp1 = load <8 x i8>* %A
+ %tmp2 = extractelement <8 x i8> %tmp1, i32 %B
+ %tmp3 = sext i8 %tmp2 to i32
+ ret i32 %tmp3
+}
+
+define i32 @vgetQ_variable_lanei32(<4 x i32>* %A, i32 %B) nounwind {
+ %tmp1 = load <4 x i32>* %A
+ %tmp2 = add <4 x i32> %tmp1, %tmp1
+ %tmp3 = extractelement <4 x i32> %tmp2, i32 %B
+ ret i32 %tmp3
+}