RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-30 15:32:52 +01:00
Code Issues Projects Releases Wiki Activity
cd7bc94022
llvm-mirror/test/CodeGen/ARM/vector-DAGCombine.ll
Quentin Colombet 49190aa8d1 [ARM] Improve the instruction selection of vector loads. 
In the ARM back-end, build_vector nodes are lowered to a target specific
build_vector that uses floating point type. 
This works well, unless the inserted bitcasts survive until instruction
selection. In that case, they incur moves between integer unit and floating
point unit that may result in inefficient code.

In other words, this conversion may introduce artificial dependencies when the
code leading to the build vector cannot be completed with a floating point type.

In particular, this happens when loads are not aligned.

Before this patch, in that case, the compiler generates general purpose loads
and creates the floating point vector from them, instead of directly using the
vector unit.

The patch uses a vector friendly sequence of code when the inserted bitcasts to
floating point survived DAGCombine.

This is done by a target specific DAGCombine that changes the target specific
build_vector into a sequence of insert_vector_elt that get rid of the bitcasts.

<rdar://problem/14170854>

llvm-svn: 185587
2013-07-03 21:42:57 +00:00

187 lines
6.8 KiB
LLVM
Raw Blame History

; RUN: llc < %s -mtriple=armv7-apple-darwin | FileCheck %s
; PR7158
define i32 @test_pr7158() nounwind {
bb.nph55.bb.nph55.split_crit_edge:
br label %bb3
bb3: ; preds = %bb3, %bb.nph55.bb.nph55.split_crit_edge
br i1 undef, label %bb.i19, label %bb3
bb.i19: ; preds = %bb.i19, %bb3
%0 = insertelement <4 x float> undef, float undef, i32 3 ; <<4 x float>> [#uses=3]
%1 = fmul <4 x float> %0, %0 ; <<4 x float>> [#uses=1]
%2 = bitcast <4 x float> %1 to <2 x double> ; <<2 x double>> [#uses=0]
%3 = fmul <4 x float> %0, undef ; <<4 x float>> [#uses=0]
br label %bb.i19
}
; Check that the DAG combiner does not arbitrarily modify BUILD_VECTORs
; after legalization.
define void @test_illegal_build_vector() nounwind {
entry:
store <2 x i64> undef, <2 x i64>* undef, align 16
%0 = load <16 x i8>* undef, align 16 ; <<16 x i8>> [#uses=1]
%1 = or <16 x i8> zeroinitializer, %0 ; <<16 x i8>> [#uses=1]
store <16 x i8> %1, <16 x i8>* undef, align 16
ret void
}
; Radar 8407927: Make sure that VMOVRRD gets optimized away when the result is
; converted back to be used as a vector type.
; CHECK: test_vmovrrd_combine
define <4 x i32> @test_vmovrrd_combine() nounwind {
entry:
br i1 undef, label %bb1, label %bb2
bb1:
%0 = bitcast <2 x i64> zeroinitializer to <2 x double>
%1 = extractelement <2 x double> %0, i32 0
%2 = bitcast double %1 to i64
%3 = insertelement <1 x i64> undef, i64 %2, i32 0
; CHECK-NOT: vmov s
; CHECK: vext.8
%4 = shufflevector <1 x i64> %3, <1 x i64> undef, <2 x i32> <i32 0, i32 1>
%tmp2006.3 = bitcast <2 x i64> %4 to <16 x i8>
%5 = shufflevector <16 x i8> %tmp2006.3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19>
%tmp2004.3 = bitcast <16 x i8> %5 to <4 x i32>
br i1 undef, label %bb2, label %bb1
bb2:
%result = phi <4 x i32> [ undef, %entry ], [ %tmp2004.3, %bb1 ]
ret <4 x i32> %result
}
; Test trying to do a ShiftCombine on illegal types.
; The vector should be split first.
define void @lshrIllegalType(<8 x i32>* %A) nounwind {
%tmp1 = load <8 x i32>* %A
%tmp2 = lshr <8 x i32> %tmp1, < i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
store <8 x i32> %tmp2, <8 x i32>* %A
ret void
}
; Test folding a binary vector operation with constant BUILD_VECTOR
; operands with i16 elements.
define void @test_i16_constant_fold() nounwind optsize {
entry:
%0 = sext <4 x i1> zeroinitializer to <4 x i16>
%1 = add <4 x i16> %0, zeroinitializer
%2 = shufflevector <4 x i16> %1, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%3 = add <8 x i16> %2, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
%4 = trunc <8 x i16> %3 to <8 x i8>
tail call void @llvm.arm.neon.vst1.v8i8(i8* undef, <8 x i8> %4, i32 1)
unreachable
}
declare void @llvm.arm.neon.vst1.v8i8(i8*, <8 x i8>, i32) nounwind
; Test that loads and stores of i64 vector elements are handled as f64 values
; so they are not split up into i32 values. Radar 8755338.
define void @i64_buildvector(i64* %ptr, <2 x i64>* %vp) nounwind {
; CHECK: i64_buildvector
; CHECK: vldr
%t0 = load i64* %ptr, align 4
%t1 = insertelement <2 x i64> undef, i64 %t0, i32 0
store <2 x i64> %t1, <2 x i64>* %vp
ret void
}
define void @i64_insertelement(i64* %ptr, <2 x i64>* %vp) nounwind {
; CHECK: i64_insertelement
; CHECK: vldr
%t0 = load i64* %ptr, align 4
%vec = load <2 x i64>* %vp
%t1 = insertelement <2 x i64> %vec, i64 %t0, i32 0
store <2 x i64> %t1, <2 x i64>* %vp
ret void
}
define void @i64_extractelement(i64* %ptr, <2 x i64>* %vp) nounwind {
; CHECK: i64_extractelement
; CHECK: vstr
%vec = load <2 x i64>* %vp
%t1 = extractelement <2 x i64> %vec, i32 0
store i64 %t1, i64* %ptr
ret void
}
; Test trying to do a AND Combine on illegal types.
define void @andVec(<3 x i8>* %A) nounwind {
%tmp = load <3 x i8>* %A, align 4
%and = and <3 x i8> %tmp, <i8 7, i8 7, i8 7>
store <3 x i8> %and, <3 x i8>* %A
ret void
}
; Test trying to do an OR Combine on illegal types.
define void @orVec(<3 x i8>* %A) nounwind {
%tmp = load <3 x i8>* %A, align 4
%or = or <3 x i8> %tmp, <i8 7, i8 7, i8 7>
store <3 x i8> %or, <3 x i8>* %A
ret void
}
; The following test was hitting an assertion in the DAG combiner when
; constant folding the multiply because the "sext undef" was translated to
; a BUILD_VECTOR with i32 0 operands, which did not match the i16 operands
; of the other BUILD_VECTOR.
define i16 @foldBuildVectors() {
%1 = sext <8 x i8> undef to <8 x i16>
%2 = mul <8 x i16> %1, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
%3 = extractelement <8 x i16> %2, i32 0
ret i16 %3
}
; Test that we are generating vrev and vext for reverse shuffles of v8i16
; shuffles.
; CHECK: reverse_v8i16
define void @reverse_v8i16(<8 x i16>* %loadaddr, <8 x i16>* %storeaddr) {
%v0 = load <8 x i16>* %loadaddr
; CHECK: vrev64.16
; CHECK: vext.16
%v1 = shufflevector <8 x i16> %v0, <8 x i16> undef,
<8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
store <8 x i16> %v1, <8 x i16>* %storeaddr
ret void
}
; Test that we are generating vrev and vext for reverse shuffles of v16i8
; shuffles.
; CHECK: reverse_v16i8
define void @reverse_v16i8(<16 x i8>* %loadaddr, <16 x i8>* %storeaddr) {
%v0 = load <16 x i8>* %loadaddr
; CHECK: vrev64.8
; CHECK: vext.8
%v1 = shufflevector <16 x i8> %v0, <16 x i8> undef,
<16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8,
i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
store <16 x i8> %v1, <16 x i8>* %storeaddr
ret void
}
; <rdar://problem/14170854>.
; vldr cannot handle unaligned loads.
; Fall back to vld1.32, which can, instead of using the general purpose loads
; followed by a costly sequence of instructions to build the vector register.
; CHECK: t3
; CHECK: vld1.32 {[[REG:d[0-9]+]][0]}
; CHECK: vld1.32 {[[REG]][1]}
; CHECK: vmull.u8 q{{[0-9]+}}, [[REG]], [[REG]]
define <8 x i16> @t3(i8 zeroext %xf, i8* nocapture %sp0, i8* nocapture %sp1, i32* nocapture %outp) {
entry:
%pix_sp0.0.cast = bitcast i8* %sp0 to i32*
%pix_sp0.0.copyload = load i32* %pix_sp0.0.cast, align 1
%pix_sp1.0.cast = bitcast i8* %sp1 to i32*
%pix_sp1.0.copyload = load i32* %pix_sp1.0.cast, align 1
%vecinit = insertelement <2 x i32> undef, i32 %pix_sp0.0.copyload, i32 0
%vecinit1 = insertelement <2 x i32> %vecinit, i32 %pix_sp1.0.copyload, i32 1
%0 = bitcast <2 x i32> %vecinit1 to <8 x i8>
%vmull.i = tail call <8 x i16> @llvm.arm.neon.vmullu.v8i16(<8 x i8> %0, <8 x i8> %0)
ret <8 x i16> %vmull.i
}
; Function Attrs: nounwind readnone
declare <8 x i16> @llvm.arm.neon.vmullu.v8i16(<8 x i8>, <8 x i8>)
Reference in New Issue View Git Blame Copy Permalink