Add codegen support for using post-increment NEON load/store instructions.
The vld1-lane, vld1-dup and vst1-lane instructions do not yet support using
post-increment versions, but all the rest of the NEON load/store instructions
should be handled now.
llvm-svn: 125014
diff --git a/llvm/test/CodeGen/ARM/vld3.ll b/llvm/test/CodeGen/ARM/vld3.ll
index 0e541bb..dde530f 100644
--- a/llvm/test/CodeGen/ARM/vld3.ll
+++ b/llvm/test/CodeGen/ARM/vld3.ll
@@ -33,6 +33,21 @@
ret <4 x i16> %tmp4
}
+;Check for a post-increment updating load with register increment.
+define <4 x i16> @vld3i16_update(i16** %ptr, i32 %inc) nounwind {
+;CHECK: vld3i16_update:
+;CHECK: vld3.16 {d16, d17, d18}, [r2], r1
+ %A = load i16** %ptr
+ %tmp0 = bitcast i16* %A to i8*
+ %tmp1 = call %struct.__neon_int16x4x3_t @llvm.arm.neon.vld3.v4i16(i8* %tmp0, i32 1)
+ %tmp2 = extractvalue %struct.__neon_int16x4x3_t %tmp1, 0
+ %tmp3 = extractvalue %struct.__neon_int16x4x3_t %tmp1, 2
+ %tmp4 = add <4 x i16> %tmp2, %tmp3
+ %tmp5 = getelementptr i16* %A, i32 %inc
+ store i16* %tmp5, i16** %ptr
+ ret <4 x i16> %tmp4
+}
+
define <2 x i32> @vld3i32(i32* %A) nounwind {
;CHECK: vld3i32:
;CHECK: vld3.32
@@ -103,6 +118,22 @@
ret <4 x i32> %tmp4
}
+;Check for a post-increment updating load.
+define <4 x i32> @vld3Qi32_update(i32** %ptr) nounwind {
+;CHECK: vld3Qi32_update:
+;CHECK: vld3.32 {d16, d18, d20}, [r1]!
+;CHECK: vld3.32 {d17, d19, d21}, [r1]!
+ %A = load i32** %ptr
+ %tmp0 = bitcast i32* %A to i8*
+ %tmp1 = call %struct.__neon_int32x4x3_t @llvm.arm.neon.vld3.v4i32(i8* %tmp0, i32 1)
+ %tmp2 = extractvalue %struct.__neon_int32x4x3_t %tmp1, 0
+ %tmp3 = extractvalue %struct.__neon_int32x4x3_t %tmp1, 2
+ %tmp4 = add <4 x i32> %tmp2, %tmp3
+ %tmp5 = getelementptr i32* %A, i32 12
+ store i32* %tmp5, i32** %ptr
+ ret <4 x i32> %tmp4
+}
+
define <4 x float> @vld3Qf(float* %A) nounwind {
;CHECK: vld3Qf:
;CHECK: vld3.32