shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/Transforms/SLPVectorizer/AMDGPU/external-shuffle.ll
Jeffrey Byrnes c9a87a50ae
[SLPVectorizer] Use accurate cost for external users of resize shuffles (#137419) 
When implementing the vectorization, we potentially need to add shuffles
for external users. In such cases, we may be shuffling a smaller vector
into a larger vector. When this happens `ResizeToVF` will just build a
poison padded identity vector. Then the to build the final shuffle, we
just use the `SK_InsertSubvector` mask.

This is possibly clearer by looking at the included test in
SLPVectorizer/AMDGPU/external-shuffle.ll

In the exit block we have a bunch of shuffles to glue the vectorized
tree match the `InsertElement` users. `TMP25` holds the result of
resizing the v2i16 vectorized sequence to match the `InsertElement` size
v16i16. Then `TMP26` is the final shuffle which replaces the
`InsertElement` sequence. This is just an insertsubvector.

However, when calculating the cost for these shuffles, we aren't
modelling this correctly. `ResizeToVF` will indicate to
`performExtractsShuffleAction` that we cannot use the original mask due
to the resize shuffle. The consequence is that the cost calculation uses
a different shuffle mask than what is ultimately used.

Going back to the included test, we can consider again `TMP26`. Clearly
we can see the shuffle uses a mask {0, 1, 2, 3, 16, 17, poison ..}.
However, we will currently calculate the cost with a mask {0, 1, 2, 3,
20, 21, ...} we have replaced 16 and 17 with 20 and 21 (Index + Vector
Size). Queries like BasicTTImpl::improveShuffleKindFromMask will not
recognize this as an `SK_InsertSubvector` mask, and targets which have
reduced costs for `SK_InsertSubvector` will not accurately calculate the
cost.
2025-06-17 08:14:05 -07:00

222 lines
21 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=slp-vectorizer,dce < %s | FileCheck -check-prefixes=GCN %s
; The insertelements in the exit block use the various parts of the vectorized tree. These external uses are just creating an identity vector using a sequence
; of insert elements. Since these insertelements are just recreating the same vectors that were produced during vectorization, they should not increase the cost of vectorization.
define void @phi_4(ptr addrspace(3) %inptr0, ptr addrspace(3) %inptr1, ptr %out, ptr %out1, ptr %out2, i32 %flag) {
; GCN-LABEL: define void @phi_4(
; GCN-SAME: ptr addrspace(3) [[INPTR0:%.*]], ptr addrspace(3) [[INPTR1:%.*]], ptr [[OUT:%.*]], ptr [[OUT1:%.*]], ptr [[OUT2:%.*]], i32 [[FLAG:%.*]]) #[[ATTR0:[0-9]+]] {
; GCN-NEXT: [[ENTRY:.*]]:
; GCN-NEXT: [[TMP0:%.*]] = load <2 x i16>, ptr addrspace(3) [[INPTR0]], align 8
; GCN-NEXT: [[GEP2:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 2
; GCN-NEXT: [[TMP1:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP2]], align 2
; GCN-NEXT: [[GEP4:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 4
; GCN-NEXT: [[TMP2:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP4]], align 8
; GCN-NEXT: [[GEP6:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 6
; GCN-NEXT: [[TMP3:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP6]], align 2
; GCN-NEXT: [[GEP8:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 8
; GCN-NEXT: [[TMP4:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP8]], align 8
; GCN-NEXT: [[GEP10:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 10
; GCN-NEXT: [[TMP5:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP10]], align 2
; GCN-NEXT: [[GEP12:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 12
; GCN-NEXT: [[TMP6:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP12]], align 8
; GCN-NEXT: [[GEP14:%.*]] = getelementptr i16, ptr addrspace(3) [[INPTR0]], i32 14
; GCN-NEXT: [[TMP7:%.*]] = load <2 x i16>, ptr addrspace(3) [[GEP14]], align 2
; GCN-NEXT: br label %[[DO_BODY:.*]]
; GCN: [[DO_BODY]]:
; GCN-NEXT: [[TMP8:%.*]] = phi <2 x i16> [ [[TMP0]], %[[ENTRY]] ], [ [[TMP16:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP9:%.*]] = phi <2 x i16> [ [[TMP1]], %[[ENTRY]] ], [ [[TMP17:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP10:%.*]] = phi <2 x i16> [ [[TMP2]], %[[ENTRY]] ], [ [[TMP18:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP11:%.*]] = phi <2 x i16> [ [[TMP3]], %[[ENTRY]] ], [ [[TMP19:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP12:%.*]] = phi <2 x i16> [ [[TMP4]], %[[ENTRY]] ], [ [[TMP20:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP13:%.*]] = phi <2 x i16> [ [[TMP5]], %[[ENTRY]] ], [ [[TMP21:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP14:%.*]] = phi <2 x i16> [ [[TMP6]], %[[ENTRY]] ], [ [[TMP22:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP42:%.*]] = phi <2 x i16> [ [[TMP7]], %[[ENTRY]] ], [ [[TMP23:%.*]], %[[DO_BODY]] ]
; GCN-NEXT: [[TMP16]] = load <2 x i16>, ptr addrspace(3) [[INPTR0]], align 8
; GCN-NEXT: [[TMP17]] = load <2 x i16>, ptr addrspace(3) [[GEP2]], align 2
; GCN-NEXT: [[TMP18]] = load <2 x i16>, ptr addrspace(3) [[GEP4]], align 8
; GCN-NEXT: [[TMP19]] = load <2 x i16>, ptr addrspace(3) [[GEP6]], align 2
; GCN-NEXT: [[TMP20]] = load <2 x i16>, ptr addrspace(3) [[GEP8]], align 8
; GCN-NEXT: [[TMP21]] = load <2 x i16>, ptr addrspace(3) [[GEP10]], align 2
; GCN-NEXT: [[TMP22]] = load <2 x i16>, ptr addrspace(3) [[GEP12]], align 8
; GCN-NEXT: [[TMP23]] = load <2 x i16>, ptr addrspace(3) [[GEP14]], align 2
; GCN-NEXT: [[CMP:%.*]] = icmp eq i32 [[FLAG]], 0
; GCN-NEXT: br i1 [[CMP]], label %[[EXIT:.*]], label %[[DO_BODY]]
; GCN: [[EXIT]]:
; GCN-NEXT: [[TMP24:%.*]] = shufflevector <2 x i16> [[TMP16]], <2 x i16> [[TMP17]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP25:%.*]] = shufflevector <2 x i16> [[TMP18]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP26:%.*]] = shufflevector <16 x i16> [[TMP24]], <16 x i16> [[TMP25]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP27:%.*]] = shufflevector <2 x i16> [[TMP19]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP28:%.*]] = shufflevector <16 x i16> [[TMP26]], <16 x i16> [[TMP27]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP29:%.*]] = shufflevector <2 x i16> [[TMP20]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP30:%.*]] = shufflevector <16 x i16> [[TMP28]], <16 x i16> [[TMP29]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP31:%.*]] = shufflevector <2 x i16> [[TMP21]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP32:%.*]] = shufflevector <16 x i16> [[TMP30]], <16 x i16> [[TMP31]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP33:%.*]] = shufflevector <2 x i16> [[TMP22]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP47:%.*]] = shufflevector <16 x i16> [[TMP32]], <16 x i16> [[TMP33]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 poison, i32 poison>
; GCN-NEXT: [[TMP48:%.*]] = shufflevector <2 x i16> [[TMP23]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP49:%.*]] = shufflevector <16 x i16> [[TMP47]], <16 x i16> [[TMP48]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 16, i32 17>
; GCN-NEXT: [[TMP37:%.*]] = shufflevector <2 x i16> [[TMP0]], <2 x i16> [[TMP1]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP38:%.*]] = shufflevector <2 x i16> [[TMP2]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP39:%.*]] = shufflevector <16 x i16> [[TMP37]], <16 x i16> [[TMP38]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP40:%.*]] = shufflevector <2 x i16> [[TMP3]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP41:%.*]] = shufflevector <16 x i16> [[TMP39]], <16 x i16> [[TMP40]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP57:%.*]] = shufflevector <2 x i16> [[TMP4]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP43:%.*]] = shufflevector <16 x i16> [[TMP41]], <16 x i16> [[TMP57]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP44:%.*]] = shufflevector <2 x i16> [[TMP5]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP45:%.*]] = shufflevector <16 x i16> [[TMP43]], <16 x i16> [[TMP44]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 16, i32 17, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP46:%.*]] = shufflevector <2 x i16> [[TMP6]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP58:%.*]] = shufflevector <16 x i16> [[TMP45]], <16 x i16> [[TMP46]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 poison, i32 poison>
; GCN-NEXT: [[TMP60:%.*]] = shufflevector <2 x i16> [[TMP7]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC2157:%.*]] = shufflevector <16 x i16> [[TMP58]], <16 x i16> [[TMP60]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 16, i32 17>
; GCN-NEXT: [[TMP50:%.*]] = shufflevector <2 x i16> [[TMP8]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[TMP51:%.*]] = shufflevector <2 x i16> [[TMP9]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC231:%.*]] = shufflevector <16 x i16> [[TMP50]], <16 x i16> [[TMP51]], <16 x i32> <i32 0, i32 1, i32 16, i32 17, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; GCN-NEXT: [[TMP52:%.*]] = shufflevector <2 x i16> [[TMP10]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC252:%.*]] = shufflevector <16 x i16> [[VEC231]], <16 x i16> [[TMP52]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; GCN-NEXT: [[TMP53:%.*]] = shufflevector <2 x i16> [[TMP11]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC273:%.*]] = shufflevector <16 x i16> [[VEC252]], <16 x i16> [[TMP53]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 16, i32 17, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; GCN-NEXT: [[TMP54:%.*]] = shufflevector <2 x i16> [[TMP12]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC294:%.*]] = shufflevector <16 x i16> [[VEC273]], <16 x i16> [[TMP54]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; GCN-NEXT: [[TMP55:%.*]] = shufflevector <2 x i16> [[TMP13]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC2115:%.*]] = shufflevector <16 x i16> [[VEC294]], <16 x i16> [[TMP55]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 16, i32 17, i32 12, i32 13, i32 14, i32 15>
; GCN-NEXT: [[TMP56:%.*]] = shufflevector <2 x i16> [[TMP14]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC2136:%.*]] = shufflevector <16 x i16> [[VEC2115]], <16 x i16> [[TMP56]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 14, i32 15>
; GCN-NEXT: [[TMP59:%.*]] = shufflevector <2 x i16> [[TMP42]], <2 x i16> poison, <16 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; GCN-NEXT: [[VEC2151:%.*]] = shufflevector <16 x i16> [[VEC2136]], <16 x i16> [[TMP59]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 16, i32 17>
; GCN-NEXT: store <16 x i16> [[VEC2157]], ptr [[OUT]], align 32
; GCN-NEXT: store <16 x i16> [[TMP49]], ptr [[OUT1]], align 32
; GCN-NEXT: store <16 x i16> [[VEC2151]], ptr [[OUT2]], align 32
; GCN-NEXT: ret void
;
entry:
%ele0 = load i16, ptr addrspace(3) %inptr0, align 8
%gep1 = getelementptr i16, ptr addrspace(3) %inptr0, i32 1
%ele1 = load i16, ptr addrspace(3) %gep1, align 1
%gep2 = getelementptr i16, ptr addrspace(3) %inptr0, i32 2
%ele2 = load i16, ptr addrspace(3) %gep2, align 2
%gep3 = getelementptr i16, ptr addrspace(3) %inptr0, i32 3
%ele3 = load i16, ptr addrspace(3) %gep3, align 1
%gep4 = getelementptr i16, ptr addrspace(3) %inptr0, i32 4
%ele4 = load i16, ptr addrspace(3) %gep4, align 8
%gep5 = getelementptr i16, ptr addrspace(3) %inptr0, i32 5
%ele5 = load i16, ptr addrspace(3) %gep5, align 1
%gep6 = getelementptr i16, ptr addrspace(3) %inptr0, i32 6
%ele6 = load i16, ptr addrspace(3) %gep6, align 2
%gep7 = getelementptr i16, ptr addrspace(3) %inptr0, i32 7
%ele7 = load i16, ptr addrspace(3) %gep7, align 1
%gep8 = getelementptr i16, ptr addrspace(3) %inptr0, i32 8
%ele8 = load i16, ptr addrspace(3) %gep8, align 8
%gep9 = getelementptr i16, ptr addrspace(3) %inptr0, i32 9
%ele9 = load i16, ptr addrspace(3) %gep9, align 1
%gep10 = getelementptr i16, ptr addrspace(3) %inptr0, i32 10
%ele10 = load i16, ptr addrspace(3) %gep10, align 2
%gep11 = getelementptr i16, ptr addrspace(3) %inptr0, i32 11
%ele11 = load i16, ptr addrspace(3) %gep11, align 1
%gep12 = getelementptr i16, ptr addrspace(3) %inptr0, i32 12
%ele12 = load i16, ptr addrspace(3) %gep12, align 8
%gep13 = getelementptr i16, ptr addrspace(3) %inptr0, i32 13
%ele13 = load i16, ptr addrspace(3) %gep13, align 1
%gep14 = getelementptr i16, ptr addrspace(3) %inptr0, i32 14
%ele14 = load i16, ptr addrspace(3) %gep14, align 2
%gep15 = getelementptr i16, ptr addrspace(3) %inptr0, i32 15
%ele15 = load i16, ptr addrspace(3) %gep15, align 1
br label %do.body
do.body:
%phi0 = phi i16 [ %ele0, %entry ], [ %otherele0, %do.body ]
%phi1 = phi i16 [ %ele1, %entry ], [ %otherele1, %do.body ]
%phi2 = phi i16 [ %ele2, %entry ], [ %otherele2, %do.body ]
%phi3 = phi i16 [ %ele3, %entry ], [ %otherele3, %do.body ]
%phi4 = phi i16 [ %ele4, %entry ], [ %otherele4, %do.body ]
%phi5 = phi i16 [ %ele5, %entry ], [ %otherele5, %do.body ]
%phi6 = phi i16 [ %ele6, %entry ], [ %otherele6, %do.body ]
%phi7 = phi i16 [ %ele7, %entry ], [ %otherele7, %do.body ]
%phi8 = phi i16 [ %ele8, %entry ], [ %otherele8, %do.body ]
%phi9 = phi i16 [ %ele9, %entry ], [ %otherele9, %do.body ]
%phi10 = phi i16 [ %ele10, %entry ], [ %otherele10, %do.body ]
%phi11 = phi i16 [ %ele11, %entry ], [ %otherele11, %do.body ]
%phi12 = phi i16 [ %ele12, %entry ], [ %otherele12, %do.body ]
%phi13 = phi i16 [ %ele13, %entry ], [ %otherele13, %do.body ]
%phi14 = phi i16 [ %ele14, %entry ], [ %otherele14, %do.body ]
%phi15 = phi i16 [ %ele15, %entry ], [ %otherele15, %do.body ]
%otherele0 = load i16, ptr addrspace(3) %inptr0, align 8
%otherele1 = load i16, ptr addrspace(3) %gep1, align 1
%otherele2 = load i16, ptr addrspace(3) %gep2, align 2
%otherele3 = load i16, ptr addrspace(3) %gep3, align 1
%otherele4 = load i16, ptr addrspace(3) %gep4, align 8
%otherele5 = load i16, ptr addrspace(3) %gep5, align 1
%otherele6 = load i16, ptr addrspace(3) %gep6, align 2
%otherele7 = load i16, ptr addrspace(3) %gep7, align 1
%otherele8 = load i16, ptr addrspace(3) %gep8, align 8
%otherele9 = load i16, ptr addrspace(3) %gep9, align 1
%otherele10 = load i16, ptr addrspace(3) %gep10, align 2
%otherele11 = load i16, ptr addrspace(3) %gep11, align 1
%otherele12 = load i16, ptr addrspace(3) %gep12, align 8
%otherele13 = load i16, ptr addrspace(3) %gep13, align 1
%otherele14 = load i16, ptr addrspace(3) %gep14, align 2
%otherele15 = load i16, ptr addrspace(3) %gep15, align 1
%cmp = icmp eq i32 %flag, 0
br i1 %cmp, label %exit, label %do.body
exit:
%vec00 = insertelement <16 x i16> poison, i16 %otherele0, i64 0
%vec01 = insertelement <16 x i16> %vec00, i16 %otherele1, i64 1
%vec02 = insertelement <16 x i16> %vec01, i16 %otherele2, i64 2
%vec03 = insertelement <16 x i16> %vec02, i16 %otherele3, i64 3
%vec04 = insertelement <16 x i16> %vec03, i16 %otherele4, i64 4
%vec05 = insertelement <16 x i16> %vec04, i16 %otherele5, i64 5
%vec06 = insertelement <16 x i16> %vec05, i16 %otherele6, i64 6
%vec07 = insertelement <16 x i16> %vec06, i16 %otherele7, i64 7
%vec08 = insertelement <16 x i16> %vec07, i16 %otherele8, i64 8
%vec09 = insertelement <16 x i16> %vec08, i16 %otherele9, i64 9
%vec010 = insertelement <16 x i16> %vec09, i16 %otherele10, i64 10
%vec011 = insertelement <16 x i16> %vec010, i16 %otherele11, i64 11
%vec012 = insertelement <16 x i16> %vec011, i16 %otherele12, i64 12
%vec013 = insertelement <16 x i16> %vec012, i16 %otherele13, i64 13
%vec014 = insertelement <16 x i16> %vec013, i16 %otherele14, i64 14
%vec015 = insertelement <16 x i16> %vec014, i16 %otherele15, i64 15
%vec10 = insertelement <16 x i16> poison, i16 %ele0, i64 0
%vec11 = insertelement <16 x i16> %vec10, i16 %ele1, i64 1
%vec12 = insertelement <16 x i16> %vec11, i16 %ele2, i64 2
%vec13 = insertelement <16 x i16> %vec12, i16 %ele3, i64 3
%vec14 = insertelement <16 x i16> %vec13, i16 %ele4, i64 4
%vec15 = insertelement <16 x i16> %vec14, i16 %ele5, i64 5
%vec16 = insertelement <16 x i16> %vec15, i16 %ele6, i64 6
%vec17 = insertelement <16 x i16> %vec16, i16 %ele7, i64 7
%vec18 = insertelement <16 x i16> %vec17, i16 %ele8, i64 8
%vec19 = insertelement <16 x i16> %vec18, i16 %ele9, i64 9
%vec110 = insertelement <16 x i16> %vec19, i16 %ele10, i64 10
%vec111 = insertelement <16 x i16> %vec110, i16 %ele11, i64 11
%vec112 = insertelement <16 x i16> %vec111, i16 %ele12, i64 12
%vec113 = insertelement <16 x i16> %vec112, i16 %ele13, i64 13
%vec114 = insertelement <16 x i16> %vec113, i16 %ele14, i64 14
%vec115 = insertelement <16 x i16> %vec114, i16 %ele15, i64 15
%vec20 = insertelement <16 x i16> poison, i16 %phi0, i64 0
%vec21 = insertelement <16 x i16> %vec20, i16 %phi1, i64 1
%vec22 = insertelement <16 x i16> %vec21, i16 %phi2, i64 2
%vec23 = insertelement <16 x i16> %vec22, i16 %phi3, i64 3
%vec24 = insertelement <16 x i16> %vec23, i16 %phi4, i64 4
%vec25 = insertelement <16 x i16> %vec24, i16 %phi5, i64 5
%vec26 = insertelement <16 x i16> %vec25, i16 %phi6, i64 6
%vec27 = insertelement <16 x i16> %vec26, i16 %phi7, i64 7
%vec28 = insertelement <16 x i16> %vec27, i16 %phi8, i64 8
%vec29 = insertelement <16 x i16> %vec28, i16 %phi9, i64 9
%vec210 = insertelement <16 x i16> %vec29, i16 %phi10, i64 10
%vec211 = insertelement <16 x i16> %vec210, i16 %phi11, i64 11
%vec212 = insertelement <16 x i16> %vec211, i16 %phi12, i64 12
%vec213 = insertelement <16 x i16> %vec212, i16 %phi13, i64 13
%vec214 = insertelement <16 x i16> %vec213, i16 %phi14, i64 14
%vec215 = insertelement <16 x i16> %vec214, i16 %phi15, i64 15
store <16 x i16> %vec115, ptr %out
store <16 x i16> %vec015, ptr %out1
store <16 x i16> %vec215, ptr %out2
ret void
}
Reference in New Issue View Git Blame Copy Permalink