[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
2018-05-06 05:19:59 +08:00
|
|
|
# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+sse2 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=ALL --check-prefix=NOT_AVX2 --check-prefix=SSE2
|
|
|
|
# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+avx -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=ALL --check-prefix=NOT_AVX2 --check-prefix=AVX1
|
|
|
|
# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+avx2 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=ALL --check-prefix=AVX2
|
2017-05-18 19:10:56 +08:00
|
|
|
|
|
|
|
--- |
|
|
|
|
define void @test_add_v32i8() {
|
|
|
|
%ret = add <32 x i8> undef, undef
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @test_add_v16i16() {
|
|
|
|
%ret = add <16 x i16> undef, undef
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @test_add_v8i32() {
|
|
|
|
%ret = add <8 x i32> undef, undef
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
define void @test_add_v4i64() {
|
|
|
|
%ret = add <4 x i64> undef, undef
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
...
|
|
|
|
---
|
|
|
|
name: test_add_v32i8
|
[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing
Summary:
This catches malformed mir files which specify alignment as log2 instead of pow2.
See https://reviews.llvm.org/D65945 for reference,
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790
Reviewers: courbet
Subscribers: MatzeB, qcolombet, dschuff, arsenm, sdardis, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, s.egerton, pzheng, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D67433
llvm-svn: 371608
2019-09-11 19:16:48 +08:00
|
|
|
alignment: 16
|
2017-05-18 19:10:56 +08:00
|
|
|
legalized: false
|
|
|
|
regBankSelected: false
|
|
|
|
registers:
|
|
|
|
- { id: 0, class: _ }
|
|
|
|
- { id: 1, class: _ }
|
|
|
|
- { id: 2, class: _ }
|
|
|
|
body: |
|
|
|
|
bb.1 (%ir-block.0):
|
2018-02-01 06:04:26 +08:00
|
|
|
liveins: $ymm0, $ymm1
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2-LABEL: name: test_add_v32i8
|
|
|
|
; SSE2: [[DEF:%[0-9]+]]:_(<32 x s8>) = IMPLICIT_DEF
|
|
|
|
; SSE2: [[DEF1:%[0-9]+]]:_(<32 x s8>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; SSE2: [[UV:%[0-9]+]]:_(<16 x s8>), [[UV1:%[0-9]+]]:_(<16 x s8>) = G_UNMERGE_VALUES [[DEF]](<32 x s8>)
|
|
|
|
; SSE2: [[UV2:%[0-9]+]]:_(<16 x s8>), [[UV3:%[0-9]+]]:_(<16 x s8>) = G_UNMERGE_VALUES [[DEF1]](<32 x s8>)
|
|
|
|
; SSE2: [[ADD:%[0-9]+]]:_(<16 x s8>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; SSE2: [[ADD1:%[0-9]+]]:_(<16 x s8>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2: [[CONCAT_VECTORS:%[0-9]+]]:_(<32 x s8>) = G_CONCAT_VECTORS [[ADD]](<16 x s8>), [[ADD1]](<16 x s8>)
|
|
|
|
; SSE2: $ymm0 = COPY [[CONCAT_VECTORS]](<32 x s8>)
|
|
|
|
; SSE2: RET 0
|
|
|
|
; AVX1-LABEL: name: test_add_v32i8
|
|
|
|
; AVX1: [[DEF:%[0-9]+]]:_(<32 x s8>) = IMPLICIT_DEF
|
|
|
|
; AVX1: [[DEF1:%[0-9]+]]:_(<32 x s8>) = IMPLICIT_DEF
|
|
|
|
; AVX1: [[UV:%[0-9]+]]:_(<16 x s8>), [[UV1:%[0-9]+]]:_(<16 x s8>) = G_UNMERGE_VALUES [[DEF]](<32 x s8>)
|
|
|
|
; AVX1: [[UV2:%[0-9]+]]:_(<16 x s8>), [[UV3:%[0-9]+]]:_(<16 x s8>) = G_UNMERGE_VALUES [[DEF1]](<32 x s8>)
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX1: [[ADD:%[0-9]+]]:_(<16 x s8>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; AVX1: [[ADD1:%[0-9]+]]:_(<16 x s8>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX1: [[CONCAT_VECTORS:%[0-9]+]]:_(<32 x s8>) = G_CONCAT_VECTORS [[ADD]](<16 x s8>), [[ADD1]](<16 x s8>)
|
|
|
|
; AVX1: $ymm0 = COPY [[CONCAT_VECTORS]](<32 x s8>)
|
|
|
|
; AVX1: RET 0
|
|
|
|
; AVX2-LABEL: name: test_add_v32i8
|
|
|
|
; AVX2: [[DEF:%[0-9]+]]:_(<32 x s8>) = IMPLICIT_DEF
|
|
|
|
; AVX2: [[DEF1:%[0-9]+]]:_(<32 x s8>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX2: [[ADD:%[0-9]+]]:_(<32 x s8>) = G_ADD [[DEF]], [[DEF1]]
|
2018-02-01 06:04:26 +08:00
|
|
|
; AVX2: $ymm0 = COPY [[ADD]](<32 x s8>)
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX2: RET 0
|
2017-05-18 19:10:56 +08:00
|
|
|
%0(<32 x s8>) = IMPLICIT_DEF
|
|
|
|
%1(<32 x s8>) = IMPLICIT_DEF
|
|
|
|
%2(<32 x s8>) = G_ADD %0, %1
|
2018-02-01 06:04:26 +08:00
|
|
|
$ymm0 = COPY %2
|
2017-05-18 19:10:56 +08:00
|
|
|
RET 0
|
|
|
|
|
|
|
|
...
|
|
|
|
---
|
|
|
|
name: test_add_v16i16
|
[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing
Summary:
This catches malformed mir files which specify alignment as log2 instead of pow2.
See https://reviews.llvm.org/D65945 for reference,
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790
Reviewers: courbet
Subscribers: MatzeB, qcolombet, dschuff, arsenm, sdardis, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, s.egerton, pzheng, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D67433
llvm-svn: 371608
2019-09-11 19:16:48 +08:00
|
|
|
alignment: 16
|
2017-05-18 19:10:56 +08:00
|
|
|
legalized: false
|
|
|
|
regBankSelected: false
|
|
|
|
registers:
|
|
|
|
- { id: 0, class: _ }
|
|
|
|
- { id: 1, class: _ }
|
|
|
|
- { id: 2, class: _ }
|
|
|
|
body: |
|
|
|
|
bb.1 (%ir-block.0):
|
2018-02-01 06:04:26 +08:00
|
|
|
liveins: $ymm0, $ymm1
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2-LABEL: name: test_add_v16i16
|
|
|
|
; SSE2: [[DEF:%[0-9]+]]:_(<16 x s16>) = IMPLICIT_DEF
|
|
|
|
; SSE2: [[DEF1:%[0-9]+]]:_(<16 x s16>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; SSE2: [[UV:%[0-9]+]]:_(<8 x s16>), [[UV1:%[0-9]+]]:_(<8 x s16>) = G_UNMERGE_VALUES [[DEF]](<16 x s16>)
|
|
|
|
; SSE2: [[UV2:%[0-9]+]]:_(<8 x s16>), [[UV3:%[0-9]+]]:_(<8 x s16>) = G_UNMERGE_VALUES [[DEF1]](<16 x s16>)
|
|
|
|
; SSE2: [[ADD:%[0-9]+]]:_(<8 x s16>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; SSE2: [[ADD1:%[0-9]+]]:_(<8 x s16>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2: [[CONCAT_VECTORS:%[0-9]+]]:_(<16 x s16>) = G_CONCAT_VECTORS [[ADD]](<8 x s16>), [[ADD1]](<8 x s16>)
|
|
|
|
; SSE2: $ymm0 = COPY [[CONCAT_VECTORS]](<16 x s16>)
|
|
|
|
; SSE2: RET 0
|
|
|
|
; AVX1-LABEL: name: test_add_v16i16
|
|
|
|
; AVX1: [[DEF:%[0-9]+]]:_(<16 x s16>) = IMPLICIT_DEF
|
|
|
|
; AVX1: [[DEF1:%[0-9]+]]:_(<16 x s16>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX1: [[UV:%[0-9]+]]:_(<8 x s16>), [[UV1:%[0-9]+]]:_(<8 x s16>) = G_UNMERGE_VALUES [[DEF]](<16 x s16>)
|
|
|
|
; AVX1: [[UV2:%[0-9]+]]:_(<8 x s16>), [[UV3:%[0-9]+]]:_(<8 x s16>) = G_UNMERGE_VALUES [[DEF1]](<16 x s16>)
|
|
|
|
; AVX1: [[ADD:%[0-9]+]]:_(<8 x s16>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; AVX1: [[ADD1:%[0-9]+]]:_(<8 x s16>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX1: [[CONCAT_VECTORS:%[0-9]+]]:_(<16 x s16>) = G_CONCAT_VECTORS [[ADD]](<8 x s16>), [[ADD1]](<8 x s16>)
|
|
|
|
; AVX1: $ymm0 = COPY [[CONCAT_VECTORS]](<16 x s16>)
|
|
|
|
; AVX1: RET 0
|
|
|
|
; AVX2-LABEL: name: test_add_v16i16
|
|
|
|
; AVX2: [[DEF:%[0-9]+]]:_(<16 x s16>) = IMPLICIT_DEF
|
|
|
|
; AVX2: [[DEF1:%[0-9]+]]:_(<16 x s16>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX2: [[ADD:%[0-9]+]]:_(<16 x s16>) = G_ADD [[DEF]], [[DEF1]]
|
2018-02-01 06:04:26 +08:00
|
|
|
; AVX2: $ymm0 = COPY [[ADD]](<16 x s16>)
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX2: RET 0
|
2017-05-18 19:10:56 +08:00
|
|
|
%0(<16 x s16>) = IMPLICIT_DEF
|
|
|
|
%1(<16 x s16>) = IMPLICIT_DEF
|
|
|
|
%2(<16 x s16>) = G_ADD %0, %1
|
2018-02-01 06:04:26 +08:00
|
|
|
$ymm0 = COPY %2
|
2017-05-18 19:10:56 +08:00
|
|
|
RET 0
|
|
|
|
|
|
|
|
...
|
|
|
|
---
|
|
|
|
name: test_add_v8i32
|
[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing
Summary:
This catches malformed mir files which specify alignment as log2 instead of pow2.
See https://reviews.llvm.org/D65945 for reference,
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790
Reviewers: courbet
Subscribers: MatzeB, qcolombet, dschuff, arsenm, sdardis, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, s.egerton, pzheng, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D67433
llvm-svn: 371608
2019-09-11 19:16:48 +08:00
|
|
|
alignment: 16
|
2017-05-18 19:10:56 +08:00
|
|
|
legalized: false
|
|
|
|
regBankSelected: false
|
|
|
|
registers:
|
|
|
|
- { id: 0, class: _ }
|
|
|
|
- { id: 1, class: _ }
|
|
|
|
- { id: 2, class: _ }
|
|
|
|
body: |
|
|
|
|
bb.1 (%ir-block.0):
|
2018-02-01 06:04:26 +08:00
|
|
|
liveins: $ymm0, $ymm1
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2-LABEL: name: test_add_v8i32
|
|
|
|
; SSE2: [[DEF:%[0-9]+]]:_(<8 x s32>) = IMPLICIT_DEF
|
|
|
|
; SSE2: [[DEF1:%[0-9]+]]:_(<8 x s32>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; SSE2: [[UV:%[0-9]+]]:_(<4 x s32>), [[UV1:%[0-9]+]]:_(<4 x s32>) = G_UNMERGE_VALUES [[DEF]](<8 x s32>)
|
|
|
|
; SSE2: [[UV2:%[0-9]+]]:_(<4 x s32>), [[UV3:%[0-9]+]]:_(<4 x s32>) = G_UNMERGE_VALUES [[DEF1]](<8 x s32>)
|
|
|
|
; SSE2: [[ADD:%[0-9]+]]:_(<4 x s32>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; SSE2: [[ADD1:%[0-9]+]]:_(<4 x s32>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[ADD]](<4 x s32>), [[ADD1]](<4 x s32>)
|
|
|
|
; SSE2: $ymm0 = COPY [[CONCAT_VECTORS]](<8 x s32>)
|
|
|
|
; SSE2: RET 0
|
|
|
|
; AVX1-LABEL: name: test_add_v8i32
|
|
|
|
; AVX1: [[DEF:%[0-9]+]]:_(<8 x s32>) = IMPLICIT_DEF
|
|
|
|
; AVX1: [[DEF1:%[0-9]+]]:_(<8 x s32>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX1: [[UV:%[0-9]+]]:_(<4 x s32>), [[UV1:%[0-9]+]]:_(<4 x s32>) = G_UNMERGE_VALUES [[DEF]](<8 x s32>)
|
|
|
|
; AVX1: [[UV2:%[0-9]+]]:_(<4 x s32>), [[UV3:%[0-9]+]]:_(<4 x s32>) = G_UNMERGE_VALUES [[DEF1]](<8 x s32>)
|
|
|
|
; AVX1: [[ADD:%[0-9]+]]:_(<4 x s32>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; AVX1: [[ADD1:%[0-9]+]]:_(<4 x s32>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX1: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s32>) = G_CONCAT_VECTORS [[ADD]](<4 x s32>), [[ADD1]](<4 x s32>)
|
|
|
|
; AVX1: $ymm0 = COPY [[CONCAT_VECTORS]](<8 x s32>)
|
|
|
|
; AVX1: RET 0
|
|
|
|
; AVX2-LABEL: name: test_add_v8i32
|
|
|
|
; AVX2: [[DEF:%[0-9]+]]:_(<8 x s32>) = IMPLICIT_DEF
|
|
|
|
; AVX2: [[DEF1:%[0-9]+]]:_(<8 x s32>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX2: [[ADD:%[0-9]+]]:_(<8 x s32>) = G_ADD [[DEF]], [[DEF1]]
|
2018-02-01 06:04:26 +08:00
|
|
|
; AVX2: $ymm0 = COPY [[ADD]](<8 x s32>)
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX2: RET 0
|
2017-05-18 19:10:56 +08:00
|
|
|
%0(<8 x s32>) = IMPLICIT_DEF
|
|
|
|
%1(<8 x s32>) = IMPLICIT_DEF
|
|
|
|
%2(<8 x s32>) = G_ADD %0, %1
|
2018-02-01 06:04:26 +08:00
|
|
|
$ymm0 = COPY %2
|
2017-05-18 19:10:56 +08:00
|
|
|
RET 0
|
|
|
|
|
|
|
|
...
|
|
|
|
---
|
|
|
|
name: test_add_v4i64
|
[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing
Summary:
This catches malformed mir files which specify alignment as log2 instead of pow2.
See https://reviews.llvm.org/D65945 for reference,
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790
Reviewers: courbet
Subscribers: MatzeB, qcolombet, dschuff, arsenm, sdardis, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, s.egerton, pzheng, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D67433
llvm-svn: 371608
2019-09-11 19:16:48 +08:00
|
|
|
alignment: 16
|
2017-05-18 19:10:56 +08:00
|
|
|
legalized: false
|
|
|
|
regBankSelected: false
|
|
|
|
registers:
|
|
|
|
- { id: 0, class: _ }
|
|
|
|
- { id: 1, class: _ }
|
|
|
|
- { id: 2, class: _ }
|
|
|
|
body: |
|
|
|
|
bb.1 (%ir-block.0):
|
2018-02-01 06:04:26 +08:00
|
|
|
liveins: $ymm0, $ymm1
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2-LABEL: name: test_add_v4i64
|
|
|
|
; SSE2: [[DEF:%[0-9]+]]:_(<4 x s64>) = IMPLICIT_DEF
|
|
|
|
; SSE2: [[DEF1:%[0-9]+]]:_(<4 x s64>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; SSE2: [[UV:%[0-9]+]]:_(<2 x s64>), [[UV1:%[0-9]+]]:_(<2 x s64>) = G_UNMERGE_VALUES [[DEF]](<4 x s64>)
|
|
|
|
; SSE2: [[UV2:%[0-9]+]]:_(<2 x s64>), [[UV3:%[0-9]+]]:_(<2 x s64>) = G_UNMERGE_VALUES [[DEF1]](<4 x s64>)
|
|
|
|
; SSE2: [[ADD:%[0-9]+]]:_(<2 x s64>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; SSE2: [[ADD1:%[0-9]+]]:_(<2 x s64>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; SSE2: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s64>) = G_CONCAT_VECTORS [[ADD]](<2 x s64>), [[ADD1]](<2 x s64>)
|
|
|
|
; SSE2: $ymm0 = COPY [[CONCAT_VECTORS]](<4 x s64>)
|
|
|
|
; SSE2: RET 0
|
|
|
|
; AVX1-LABEL: name: test_add_v4i64
|
|
|
|
; AVX1: [[DEF:%[0-9]+]]:_(<4 x s64>) = IMPLICIT_DEF
|
|
|
|
; AVX1: [[DEF1:%[0-9]+]]:_(<4 x s64>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX1: [[UV:%[0-9]+]]:_(<2 x s64>), [[UV1:%[0-9]+]]:_(<2 x s64>) = G_UNMERGE_VALUES [[DEF]](<4 x s64>)
|
|
|
|
; AVX1: [[UV2:%[0-9]+]]:_(<2 x s64>), [[UV3:%[0-9]+]]:_(<2 x s64>) = G_UNMERGE_VALUES [[DEF1]](<4 x s64>)
|
|
|
|
; AVX1: [[ADD:%[0-9]+]]:_(<2 x s64>) = G_ADD [[UV]], [[UV2]]
|
|
|
|
; AVX1: [[ADD1:%[0-9]+]]:_(<2 x s64>) = G_ADD [[UV1]], [[UV3]]
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX1: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s64>) = G_CONCAT_VECTORS [[ADD]](<2 x s64>), [[ADD1]](<2 x s64>)
|
|
|
|
; AVX1: $ymm0 = COPY [[CONCAT_VECTORS]](<4 x s64>)
|
|
|
|
; AVX1: RET 0
|
|
|
|
; AVX2-LABEL: name: test_add_v4i64
|
|
|
|
; AVX2: [[DEF:%[0-9]+]]:_(<4 x s64>) = IMPLICIT_DEF
|
|
|
|
; AVX2: [[DEF1:%[0-9]+]]:_(<4 x s64>) = IMPLICIT_DEF
|
[GISel]: Rework legalization algorithm for better elimination of
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
2017-11-15 06:42:19 +08:00
|
|
|
; AVX2: [[ADD:%[0-9]+]]:_(<4 x s64>) = G_ADD [[DEF]], [[DEF1]]
|
2018-02-01 06:04:26 +08:00
|
|
|
; AVX2: $ymm0 = COPY [[ADD]](<4 x s64>)
|
2019-03-01 21:13:40 +08:00
|
|
|
; AVX2: RET 0
|
2017-05-18 19:10:56 +08:00
|
|
|
%0(<4 x s64>) = IMPLICIT_DEF
|
|
|
|
%1(<4 x s64>) = IMPLICIT_DEF
|
|
|
|
%2(<4 x s64>) = G_ADD %0, %1
|
2018-02-01 06:04:26 +08:00
|
|
|
$ymm0 = COPY %2
|
2017-05-18 19:10:56 +08:00
|
|
|
RET 0
|
|
|
|
|
|
|
|
...
|