llvm-6502/test/CodeGen/ARM/vld2.ll
Bob Wilson 7d24705f65 Change register allocation order for ARM VFP and NEON registers to put the
callee-saved registers at the end of the lists.  Also prefer to avoid using
the low registers that are in register subclasses required by certain
instructions, so that those registers will more likely be available when needed.
This change makes a huge improvement in spilling in some cases.  Thanks to
Jakob for helping me realize the problem.

Most of this patch is fixing the testsuite.  There are quite a few places
where we're checking for specific registers.  I changed those to wildcards
in places where that doesn't weaken the tests.  The spill-q.ll and
thumb2-spill-q.ll tests stopped spilling with this change, so I added a bunch
of live values to force spills on those tests.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116055 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-08 06:15:13 +00:00

127 lines
5.3 KiB
LLVM

; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s
%struct.__neon_int8x8x2_t = type { <8 x i8>, <8 x i8> }
%struct.__neon_int16x4x2_t = type { <4 x i16>, <4 x i16> }
%struct.__neon_int32x2x2_t = type { <2 x i32>, <2 x i32> }
%struct.__neon_float32x2x2_t = type { <2 x float>, <2 x float> }
%struct.__neon_int64x1x2_t = type { <1 x i64>, <1 x i64> }
%struct.__neon_int8x16x2_t = type { <16 x i8>, <16 x i8> }
%struct.__neon_int16x8x2_t = type { <8 x i16>, <8 x i16> }
%struct.__neon_int32x4x2_t = type { <4 x i32>, <4 x i32> }
%struct.__neon_float32x4x2_t = type { <4 x float>, <4 x float> }
define <8 x i8> @vld2i8(i8* %A) nounwind {
;CHECK: vld2i8:
;Check the alignment value. Max for this instruction is 128 bits:
;CHECK: vld2.8 {d16, d17}, [r0, :64]
%tmp1 = call %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2.v8i8(i8* %A, i32 8)
%tmp2 = extractvalue %struct.__neon_int8x8x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x8x2_t %tmp1, 1
%tmp4 = add <8 x i8> %tmp2, %tmp3
ret <8 x i8> %tmp4
}
define <4 x i16> @vld2i16(i16* %A) nounwind {
;CHECK: vld2i16:
;Check the alignment value. Max for this instruction is 128 bits:
;CHECK: vld2.16 {d16, d17}, [r0, :128]
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x4x2_t @llvm.arm.neon.vld2.v4i16(i8* %tmp0, i32 32)
%tmp2 = extractvalue %struct.__neon_int16x4x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x4x2_t %tmp1, 1
%tmp4 = add <4 x i16> %tmp2, %tmp3
ret <4 x i16> %tmp4
}
define <2 x i32> @vld2i32(i32* %A) nounwind {
;CHECK: vld2i32:
;CHECK: vld2.32
%tmp0 = bitcast i32* %A to i8*
%tmp1 = call %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2.v2i32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_int32x2x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int32x2x2_t %tmp1, 1
%tmp4 = add <2 x i32> %tmp2, %tmp3
ret <2 x i32> %tmp4
}
define <2 x float> @vld2f(float* %A) nounwind {
;CHECK: vld2f:
;CHECK: vld2.32
%tmp0 = bitcast float* %A to i8*
%tmp1 = call %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2.v2f32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_float32x2x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_float32x2x2_t %tmp1, 1
%tmp4 = fadd <2 x float> %tmp2, %tmp3
ret <2 x float> %tmp4
}
define <1 x i64> @vld2i64(i64* %A) nounwind {
;CHECK: vld2i64:
;Check the alignment value. Max for this instruction is 128 bits:
;CHECK: vld1.64 {d16, d17}, [r0, :128]
%tmp0 = bitcast i64* %A to i8*
%tmp1 = call %struct.__neon_int64x1x2_t @llvm.arm.neon.vld2.v1i64(i8* %tmp0, i32 32)
%tmp2 = extractvalue %struct.__neon_int64x1x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int64x1x2_t %tmp1, 1
%tmp4 = add <1 x i64> %tmp2, %tmp3
ret <1 x i64> %tmp4
}
define <16 x i8> @vld2Qi8(i8* %A) nounwind {
;CHECK: vld2Qi8:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld2.8 {d16, d17, d18, d19}, [r0, :64]
%tmp1 = call %struct.__neon_int8x16x2_t @llvm.arm.neon.vld2.v16i8(i8* %A, i32 8)
%tmp2 = extractvalue %struct.__neon_int8x16x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int8x16x2_t %tmp1, 1
%tmp4 = add <16 x i8> %tmp2, %tmp3
ret <16 x i8> %tmp4
}
define <8 x i16> @vld2Qi16(i16* %A) nounwind {
;CHECK: vld2Qi16:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld2.16 {d16, d17, d18, d19}, [r0, :128]
%tmp0 = bitcast i16* %A to i8*
%tmp1 = call %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2.v8i16(i8* %tmp0, i32 16)
%tmp2 = extractvalue %struct.__neon_int16x8x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int16x8x2_t %tmp1, 1
%tmp4 = add <8 x i16> %tmp2, %tmp3
ret <8 x i16> %tmp4
}
define <4 x i32> @vld2Qi32(i32* %A) nounwind {
;CHECK: vld2Qi32:
;Check the alignment value. Max for this instruction is 256 bits:
;CHECK: vld2.32 {d16, d17, d18, d19}, [r0, :256]
%tmp0 = bitcast i32* %A to i8*
%tmp1 = call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32(i8* %tmp0, i32 64)
%tmp2 = extractvalue %struct.__neon_int32x4x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_int32x4x2_t %tmp1, 1
%tmp4 = add <4 x i32> %tmp2, %tmp3
ret <4 x i32> %tmp4
}
define <4 x float> @vld2Qf(float* %A) nounwind {
;CHECK: vld2Qf:
;CHECK: vld2.32
%tmp0 = bitcast float* %A to i8*
%tmp1 = call %struct.__neon_float32x4x2_t @llvm.arm.neon.vld2.v4f32(i8* %tmp0, i32 1)
%tmp2 = extractvalue %struct.__neon_float32x4x2_t %tmp1, 0
%tmp3 = extractvalue %struct.__neon_float32x4x2_t %tmp1, 1
%tmp4 = fadd <4 x float> %tmp2, %tmp3
ret <4 x float> %tmp4
}
declare %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2.v8i8(i8*, i32) nounwind readonly
declare %struct.__neon_int16x4x2_t @llvm.arm.neon.vld2.v4i16(i8*, i32) nounwind readonly
declare %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2.v2i32(i8*, i32) nounwind readonly
declare %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2.v2f32(i8*, i32) nounwind readonly
declare %struct.__neon_int64x1x2_t @llvm.arm.neon.vld2.v1i64(i8*, i32) nounwind readonly
declare %struct.__neon_int8x16x2_t @llvm.arm.neon.vld2.v16i8(i8*, i32) nounwind readonly
declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2.v8i16(i8*, i32) nounwind readonly
declare %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32(i8*, i32) nounwind readonly
declare %struct.__neon_float32x4x2_t @llvm.arm.neon.vld2.v4f32(i8*, i32) nounwind readonly