maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-unmerge-values.mir

1154 lines
50 KiB
Plaintext
Raw Normal View History

AMDGPU/GlobalISel: Make some G_MERGE_VALUEs legal llvm-svn: 327267
2018-03-12 21:35:43 +08:00
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
[GlobalISel] support widen unmerge if WideTy > SrcTy Summary: Widening G_UNMERGE_VALUES to a type which is larger than the original source type is the same as widening it to the same type as the source type: in both cases, G_UNMERGE_VALUES has to be replaced with bit arithmetic which. Although the arithmetic itself is independent of whether the source type is smaller or equal to the widen type, widening the source type to the widen type should result in less artifacts being emitted, since this is the type that the user explicitly requested. Reviewers: arsenm, dsanders, aemerson, aditya_nandakumar Reviewed By: arsenm, dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, hiraditya, volkan, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D76494
2020-03-20 21:46:01 +08:00
# RUN: llc -mtriple=amdgcn-- -O0 -run-pass=legalizer -o - %s | FileCheck %s
AMDGPU/GlobalISel: Make some G_MERGE_VALUEs legal llvm-svn: 327267
2018-03-12 21:35:43 +08:00
---
AMDGPU/GlobalISel: Fix some crashs in g_unmerge_values/g_merge_values This was crashing in the predicate function assuming the value is a vector. Copy more of what AArch64 uses. This probably needs more refinement later, but I don't exactly understand what it means in some cases, particularly since any legalization for these seems to be missing. llvm-svn: 351693
2019-01-21 02:40:36 +08:00
name: test_unmerge_s32_s64
AMDGPU/GlobalISel: Make some G_MERGE_VALUEs legal llvm-svn: 327267
2018-03-12 21:35:43 +08:00
body: |
bb.0:
AMDGPU/GlobalISel: Fix some crashs in g_unmerge_values/g_merge_values This was crashing in the predicate function assuming the value is a vector. Copy more of what AArch64 uses. This probably needs more refinement later, but I don't exactly understand what it means in some cases, particularly since any legalization for these seems to be missing. llvm-svn: 351693
2019-01-21 02:40:36 +08:00
; CHECK-LABEL: name: test_unmerge_s32_s64
AMDGPU/GlobalISel: Make some G_MERGE_VALUEs legal llvm-svn: 327267
2018-03-12 21:35:43 +08:00
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[C]](s64)
; CHECK: $vgpr0 = COPY [[UV]](s32)
AMDGPU/GlobalISel: Fix some crashs in g_unmerge_values/g_merge_values This was crashing in the predicate function assuming the value is a vector. Copy more of what AArch64 uses. This probably needs more refinement later, but I don't exactly understand what it means in some cases, particularly since any legalization for these seems to be missing. llvm-svn: 351693
2019-01-21 02:40:36 +08:00
; CHECK: $vgpr1 = COPY [[UV1]](s32)
AMDGPU/GlobalISel: Make some G_MERGE_VALUEs legal llvm-svn: 327267
2018-03-12 21:35:43 +08:00
%0:_(s64) = G_CONSTANT i64 0
%1:_(s32), %2:_(s32) = G_UNMERGE_VALUES %0:_(s64)
$vgpr0 = COPY %1(s32)
AMDGPU/GlobalISel: Fix some crashs in g_unmerge_values/g_merge_values This was crashing in the predicate function assuming the value is a vector. Copy more of what AArch64 uses. This probably needs more refinement later, but I don't exactly understand what it means in some cases, particularly since any legalization for these seems to be missing. llvm-svn: 351693
2019-01-21 02:40:36 +08:00
$vgpr1 = COPY %2(s32)
...
---
name: test_unmerge_s32_v2s32
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; CHECK: $vgpr0 = COPY [[UV]](s32)
; CHECK: $vgpr21 = COPY [[UV1]](s32)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(s32), %2:_(s32) = G_UNMERGE_VALUES %0
$vgpr0 = COPY %1
$vgpr21= COPY %2
...
---
name: test_unmerge_s16_v2s16
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s16_v2s16
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
AMDGPU/GlobalISel: Fix some crashs in g_unmerge_values/g_merge_values This was crashing in the predicate function assuming the value is a vector. Copy more of what AArch64 uses. This probably needs more refinement later, but I don't exactly understand what it means in some cases, particularly since any legalization for these seems to be missing. llvm-svn: 351693
2019-01-21 02:40:36 +08:00
%0:_(<2 x s16>) = COPY $vgpr0
%1:_(s16), %2:_(s16) = G_UNMERGE_VALUES %0
%3:_(s32) = G_ANYEXT %1
%4:_(s32) = G_ANYEXT %2
$vgpr0 = COPY %3
$vgpr1 = COPY %4
...
---
name: test_unmerge_s16_v3s16
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_s16_v3s16
GlobalISel: Implement moreElementsVector for implicit_def llvm-svn: 353754
2019-02-12 06:00:39 +08:00
; CHECK: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
GlobalISel: Reimplement moreElementsVectorDst Use pad with undef and unmerge with unused results. This is annoyingly similar to several other places in LegalizerHelper, but they're all slightly different.
2020-02-03 06:42:02 +08:00
; CHECK: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; CHECK: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
; CHECK: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
; CHECK: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
; CHECK: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
GlobalISel: Implement moreElementsVector for G_UNMERGE_VALUES sources This is necessary for handling <3 x s16> on AMDGPU, assuming this should be handled as 2 separate legalization actions. The alternative would be for fewerElementsVector to handle 3->2. llvm-svn: 369547
2019-08-22 00:59:10 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
GlobalISel: Reimplement moreElementsVectorDst Use pad with undef and unmerge with unused results. This is annoyingly similar to several other places in LegalizerHelper, but they're all slightly different.
2020-02-03 06:42:02 +08:00
; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>)
GlobalISel: Implement moreElementsVector for G_UNMERGE_VALUES sources This is necessary for handling <3 x s16> on AMDGPU, assuming this should be handled as 2 separate legalization actions. The alternative would be for fewerElementsVector to handle 3->2. llvm-svn: 369547
2019-08-22 00:59:10 +08:00
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: $vgpr0 = COPY [[COPY]](s32)
; CHECK: $vgpr1 = COPY [[COPY1]](s32)
; CHECK: $vgpr2 = COPY [[COPY2]](s32)
AMDGPU/GlobalISel: Fix some crashs in g_unmerge_values/g_merge_values This was crashing in the predicate function assuming the value is a vector. Copy more of what AArch64 uses. This probably needs more refinement later, but I don't exactly understand what it means in some cases, particularly since any legalization for these seems to be missing. llvm-svn: 351693
2019-01-21 02:40:36 +08:00
%0:_(<3 x s16>) = G_IMPLICIT_DEF
%1:_(s16), %2:_(s16), %3:_(s16) = G_UNMERGE_VALUES %0
%4:_(s32) = G_ANYEXT %1
%5:_(s32) = G_ANYEXT %2
%6:_(s32) = G_ANYEXT %3
$vgpr0 = COPY %4
$vgpr1 = COPY %5
$vgpr2 = COPY %6
AMDGPU/GlobalISel: Make some G_MERGE_VALUEs legal llvm-svn: 327267
2018-03-12 21:35:43 +08:00
...
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
---
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
name: test_unmerge_s16_v4s16
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s16_v4s16
; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16) = G_UNMERGE_VALUES %0
%5:_(s32) = G_ANYEXT %1
%6:_(s32) = G_ANYEXT %2
%7:_(s32) = G_ANYEXT %3
%8:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
$vgpr2 = COPY %7
$vgpr3 = COPY %8
...
---
name: test_unmerge_s16_v6s16
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s16_v6s16
; CHECK: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; CHECK: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 x s16>)
; CHECK: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
; CHECK: $vgpr4 = COPY [[COPY5]](s32)
; CHECK: $vgpr5 = COPY [[COPY6]](s32)
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
%0:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
%1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16), %5:_(s16), %6:_(s16) = G_UNMERGE_VALUES %0
%7:_(s32) = G_ANYEXT %1
%8:_(s32) = G_ANYEXT %2
%9:_(s32) = G_ANYEXT %3
%10:_(s32) = G_ANYEXT %4
%11:_(s32) = G_ANYEXT %5
%12:_(s32) = G_ANYEXT %6
$vgpr0 = COPY %7
$vgpr1 = COPY %8
$vgpr2 = COPY %9
$vgpr3 = COPY %10
$vgpr4 = COPY %11
$vgpr5 = COPY %12
...
---
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
name: test_unmerge_s8_s16
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s8_s16
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
%2:_(s8), %3:_(s8) = G_UNMERGE_VALUES %1
%4:_(s32) = G_ANYEXT %2
%5:_(s32) = G_ANYEXT %3
$vgpr0 = COPY %4
$vgpr1 = COPY %5
...
---
name: test_unmerge_s8_s32
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s8_s32
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s8), %2:_(s8), %3:_(s8), %4:_(s8) = G_UNMERGE_VALUES %0
%5:_(s32) = G_ANYEXT %1
%6:_(s32) = G_ANYEXT %2
%7:_(s32) = G_ANYEXT %3
%8:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
$vgpr2 = COPY %7
$vgpr3 = COPY %8
...
AMDGPU/GlobalISel: Add some baseline tests for unmerge legalization
2020-01-21 21:18:39 +08:00
---
name: test_unmerge_s8_s48
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s8_s48
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[AND]], [[COPY1]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C2]]
; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[AND1]], [[COPY3]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C2]]
; CHECK: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[AND2]], [[C1]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY10:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
; CHECK: $vgpr0 = COPY [[COPY6]](s32)
; CHECK: $vgpr1 = COPY [[COPY7]](s32)
; CHECK: $vgpr2 = COPY [[COPY8]](s32)
; CHECK: $vgpr3 = COPY [[COPY9]](s32)
; CHECK: $vgpr4 = COPY [[COPY10]](s32)
; CHECK: $vgpr5 = COPY [[COPY11]](s32)
AMDGPU/GlobalISel: Add some baseline tests for unmerge legalization
2020-01-21 21:18:39 +08:00
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s48) = G_TRUNC %0
%2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8), %6:_(s8), %7:_(s8) = G_UNMERGE_VALUES %1
%8:_(s32) = G_ANYEXT %2
%9:_(s32) = G_ANYEXT %3
%10:_(s32) = G_ANYEXT %4
%11:_(s32) = G_ANYEXT %5
%12:_(s32) = G_ANYEXT %6
%13:_(s32) = G_ANYEXT %7
$vgpr0 = COPY %8
$vgpr1 = COPY %9
$vgpr2 = COPY %10
$vgpr3 = COPY %11
$vgpr4 = COPY %12
$vgpr5 = COPY %13
...
---
name: test_unmerge_s16_s48
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s16_s48
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
AMDGPU/GlobalISel: Add some baseline tests for unmerge legalization
2020-01-21 21:18:39 +08:00
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s48) = G_TRUNC %0
%2:_(s16), %3:_(s16), %4:_(s16) = G_UNMERGE_VALUES %1
%5:_(s32) = G_ANYEXT %2
%6:_(s32) = G_ANYEXT %3
%7:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
$vgpr2 = COPY %7
...
---
name: test_unmerge_s8_s64
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s8_s64
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[AND]], [[COPY1]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C2]]
; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[AND1]], [[COPY3]](s32)
; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C2]]
; CHECK: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[AND2]], [[COPY5]](s32)
; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C2]]
; CHECK: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[AND3]], [[C1]](s32)
; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; CHECK: [[COPY12:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
; CHECK: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
; CHECK: $vgpr0 = COPY [[COPY8]](s32)
; CHECK: $vgpr1 = COPY [[COPY9]](s32)
; CHECK: $vgpr2 = COPY [[COPY10]](s32)
; CHECK: $vgpr3 = COPY [[COPY11]](s32)
; CHECK: $vgpr4 = COPY [[COPY12]](s32)
; CHECK: $vgpr5 = COPY [[COPY13]](s32)
; CHECK: $vgpr6 = COPY [[COPY14]](s32)
; CHECK: $vgpr7 = COPY [[COPY15]](s32)
AMDGPU/GlobalISel: Add some baseline tests for unmerge legalization
2020-01-21 21:18:39 +08:00
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s8), %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8), %6:_(s8), %7:_(s8), %8:_(s8) = G_UNMERGE_VALUES %0
%10:_(s32) = G_ANYEXT %1
%11:_(s32) = G_ANYEXT %2
%12:_(s32) = G_ANYEXT %3
%13:_(s32) = G_ANYEXT %4
%14:_(s32) = G_ANYEXT %5
%15:_(s32) = G_ANYEXT %6
%16:_(s32) = G_ANYEXT %7
%17:_(s32) = G_ANYEXT %8
$vgpr0 = COPY %10
$vgpr1 = COPY %11
$vgpr2 = COPY %12
$vgpr3 = COPY %13
$vgpr4 = COPY %14
$vgpr5 = COPY %15
$vgpr6 = COPY %16
$vgpr7 = COPY %17
...
---
name: test_unmerge_s8_p1
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s8_p1
; CHECK: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
GlobalISel: Support widening unmerge results with pointer source
2020-01-22 00:12:36 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](p1)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[AND]], [[COPY1]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C2]]
; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[AND1]], [[COPY3]](s32)
; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C2]]
; CHECK: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[AND2]], [[COPY5]](s32)
; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C2]]
; CHECK: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[AND3]], [[C1]](s32)
; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY11:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; CHECK: [[COPY12:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[COPY13:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
; CHECK: [[COPY14:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY15:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
; CHECK: $vgpr0 = COPY [[COPY8]](s32)
; CHECK: $vgpr1 = COPY [[COPY9]](s32)
; CHECK: $vgpr2 = COPY [[COPY10]](s32)
; CHECK: $vgpr3 = COPY [[COPY11]](s32)
; CHECK: $vgpr4 = COPY [[COPY12]](s32)
; CHECK: $vgpr5 = COPY [[COPY13]](s32)
; CHECK: $vgpr6 = COPY [[COPY14]](s32)
; CHECK: $vgpr7 = COPY [[COPY15]](s32)
AMDGPU/GlobalISel: Add some baseline tests for unmerge legalization
2020-01-21 21:18:39 +08:00
%0:_(p1) = COPY $vgpr0_vgpr1
%1:_(s8), %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8), %6:_(s8), %7:_(s8), %8:_(s8) = G_UNMERGE_VALUES %0
%10:_(s32) = G_ANYEXT %1
%11:_(s32) = G_ANYEXT %2
%12:_(s32) = G_ANYEXT %3
%13:_(s32) = G_ANYEXT %4
%14:_(s32) = G_ANYEXT %5
%15:_(s32) = G_ANYEXT %6
%16:_(s32) = G_ANYEXT %7
%17:_(s32) = G_ANYEXT %8
$vgpr0 = COPY %10
$vgpr1 = COPY %11
$vgpr2 = COPY %12
$vgpr3 = COPY %13
$vgpr4 = COPY %14
$vgpr5 = COPY %15
$vgpr6 = COPY %16
$vgpr7 = COPY %17
...
GlobalISel: Support widening unmerge results with pointer source
2020-01-22 00:12:36 +08:00
---
name: test_unmerge_s4_p1
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s4_p1
; CHECK: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](p1)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[AND]], [[COPY1]](s32)
; CHECK: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[C3]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C2]]
; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[AND1]], [[COPY3]](s32)
; CHECK: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[C4]](s32)
; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C2]]
; CHECK: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[AND2]], [[COPY5]](s32)
; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY8]], [[C2]]
; CHECK: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[AND3]], [[COPY7]](s32)
; CHECK: [[COPY9:%[0-9]+]]:_(s32) = COPY [[C3]](s32)
; CHECK: [[COPY10:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND4:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C2]]
; CHECK: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[AND4]], [[COPY9]](s32)
; CHECK: [[COPY11:%[0-9]+]]:_(s32) = COPY [[C4]](s32)
; CHECK: [[COPY12:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND5:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C2]]
; CHECK: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[AND5]], [[COPY11]](s32)
; CHECK: [[COPY13:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
; CHECK: [[COPY14:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[AND6:%[0-9]+]]:_(s32) = G_AND [[COPY14]], [[C2]]
; CHECK: [[LSHR8:%[0-9]+]]:_(s32) = G_LSHR [[AND6]], [[COPY13]](s32)
; CHECK: [[COPY15:%[0-9]+]]:_(s32) = COPY [[C3]](s32)
; CHECK: [[COPY16:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[AND7:%[0-9]+]]:_(s32) = G_AND [[COPY16]], [[C2]]
; CHECK: [[LSHR9:%[0-9]+]]:_(s32) = G_LSHR [[AND7]], [[COPY15]](s32)
; CHECK: [[COPY17:%[0-9]+]]:_(s32) = COPY [[C4]](s32)
; CHECK: [[COPY18:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[AND8:%[0-9]+]]:_(s32) = G_AND [[COPY18]], [[C2]]
; CHECK: [[LSHR10:%[0-9]+]]:_(s32) = G_LSHR [[AND8]], [[COPY17]](s32)
; CHECK: [[COPY19:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[AND9:%[0-9]+]]:_(s32) = G_AND [[COPY19]], [[C2]]
; CHECK: [[LSHR11:%[0-9]+]]:_(s32) = G_LSHR [[AND9]], [[C1]](s32)
; CHECK: [[COPY20:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[AND10:%[0-9]+]]:_(s32) = G_AND [[COPY20]], [[C2]]
; CHECK: [[LSHR12:%[0-9]+]]:_(s32) = G_LSHR [[AND10]], [[C3]](s32)
; CHECK: [[COPY21:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[AND11:%[0-9]+]]:_(s32) = G_AND [[COPY21]], [[C2]]
; CHECK: [[LSHR13:%[0-9]+]]:_(s32) = G_LSHR [[AND11]], [[C4]](s32)
; CHECK: [[COPY22:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[COPY23:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: [[COPY24:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; CHECK: [[COPY25:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
; CHECK: [[COPY26:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY27:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
; CHECK: [[COPY28:%[0-9]+]]:_(s32) = COPY [[LSHR6]](s32)
; CHECK: [[COPY29:%[0-9]+]]:_(s32) = COPY [[LSHR7]](s32)
; CHECK: [[COPY30:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[COPY31:%[0-9]+]]:_(s32) = COPY [[LSHR8]](s32)
; CHECK: [[COPY32:%[0-9]+]]:_(s32) = COPY [[LSHR9]](s32)
; CHECK: [[COPY33:%[0-9]+]]:_(s32) = COPY [[LSHR10]](s32)
; CHECK: [[COPY34:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY35:%[0-9]+]]:_(s32) = COPY [[LSHR11]](s32)
; CHECK: [[COPY36:%[0-9]+]]:_(s32) = COPY [[LSHR12]](s32)
; CHECK: [[COPY37:%[0-9]+]]:_(s32) = COPY [[LSHR13]](s32)
; CHECK: $vgpr0 = COPY [[COPY22]](s32)
; CHECK: $vgpr1 = COPY [[COPY23]](s32)
; CHECK: $vgpr2 = COPY [[COPY24]](s32)
; CHECK: $vgpr3 = COPY [[COPY25]](s32)
; CHECK: $vgpr4 = COPY [[COPY26]](s32)
; CHECK: $vgpr5 = COPY [[COPY27]](s32)
; CHECK: $vgpr6 = COPY [[COPY28]](s32)
; CHECK: $vgpr7 = COPY [[COPY29]](s32)
; CHECK: $vgpr8 = COPY [[COPY30]](s32)
; CHECK: $vgpr9 = COPY [[COPY31]](s32)
; CHECK: $vgpr10 = COPY [[COPY32]](s32)
; CHECK: $vgpr11 = COPY [[COPY33]](s32)
; CHECK: $vgpr12 = COPY [[COPY34]](s32)
; CHECK: $vgpr13 = COPY [[COPY35]](s32)
; CHECK: $vgpr14 = COPY [[COPY36]](s32)
; CHECK: $vgpr15 = COPY [[COPY37]](s32)
%0:_(p1) = COPY $vgpr0_vgpr1
%1:_(s4), %2:_(s4), %3:_(s4), %4:_(s4), %5:_(s4), %6:_(s4), %7:_(s4), %8:_(s4), %9:_(s4), %10:_(s4), %11:_(s4), %12:_(s4), %13:_(s4), %14:_(s4), %15:_(s4), %16:_(s4) = G_UNMERGE_VALUES %0
%17:_(s32) = G_ANYEXT %1
%18:_(s32) = G_ANYEXT %2
%19:_(s32) = G_ANYEXT %3
%20:_(s32) = G_ANYEXT %4
%21:_(s32) = G_ANYEXT %5
%22:_(s32) = G_ANYEXT %6
%23:_(s32) = G_ANYEXT %7
%24:_(s32) = G_ANYEXT %8
%25:_(s32) = G_ANYEXT %9
%26:_(s32) = G_ANYEXT %10
%27:_(s32) = G_ANYEXT %11
%28:_(s32) = G_ANYEXT %12
%29:_(s32) = G_ANYEXT %13
%30:_(s32) = G_ANYEXT %14
%31:_(s32) = G_ANYEXT %15
%32:_(s32) = G_ANYEXT %16
$vgpr0 = COPY %17
$vgpr1 = COPY %18
$vgpr2 = COPY %19
$vgpr3 = COPY %20
$vgpr4 = COPY %21
$vgpr5 = COPY %22
$vgpr6 = COPY %23
$vgpr7 = COPY %24
$vgpr8 = COPY %25
$vgpr9 = COPY %26
$vgpr10 = COPY %27
$vgpr11 = COPY %28
$vgpr12 = COPY %29
$vgpr13 = COPY %30
$vgpr14 = COPY %31
$vgpr15 = COPY %32
...
---
name: test_unmerge_s16_p1
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s16_p1
; CHECK: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](p1)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
%0:_(p1) = COPY $vgpr0_vgpr1
%1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16) = G_UNMERGE_VALUES %0
%5:_(s32) = G_ANYEXT %1
%6:_(s32) = G_ANYEXT %2
%7:_(s32) = G_ANYEXT %3
%8:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
$vgpr2 = COPY %7
$vgpr3 = COPY %8
...
---
name: test_unmerge_s32_p1
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s32_p1
; CHECK: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](p1)
; CHECK: $vgpr0 = COPY [[UV]](s32)
; CHECK: $vgpr1 = COPY [[UV1]](s32)
%0:_(p1) = COPY $vgpr0_vgpr1
%1:_(s32), %2:_(s32) = G_UNMERGE_VALUES %0
$vgpr0 = COPY %1
$vgpr1 = COPY %2
...
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
---
name: test_unmerge_s16_s32
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s16_s32
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s16), %2:_(s16) = G_UNMERGE_VALUES %0
GlobalISel: Support widening unmerge results with pointer source
2020-01-22 00:12:36 +08:00
%3:_(s32) = G_ANYEXT %1
%4:_(s32) = G_ANYEXT %2
$vgpr0 = COPY %3
$vgpr1 = COPY %4
...
---
name: test_unmerge_s16_p3
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s16_p3
; CHECK: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
; CHECK: [[PTRTOINT:%[0-9]+]]:_(s32) = G_PTRTOINT [[COPY]](p3)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[PTRTOINT]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[PTRTOINT]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
%0:_(p3) = COPY $vgpr0
%1:_(s16), %2:_(s16) = G_UNMERGE_VALUES %0
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
%3:_(s32) = G_ANYEXT %1
%4:_(s32) = G_ANYEXT %2
$vgpr0 = COPY %3
$vgpr1 = COPY %4
...
AMDGPU/GlobalISel: Add a testcase for G_UNMERGE_VALUES legalization I had a note that this doesn't work, but it seems to now.
2020-03-25 09:52:03 +08:00
---
name: test_unmerge_s8_p3
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s8_p3
; CHECK: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
; CHECK: [[PTRTOINT:%[0-9]+]]:_(s32) = G_PTRTOINT [[COPY]](p3)
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
AMDGPU/GlobalISel: Add a testcase for G_UNMERGE_VALUES legalization I had a note that this doesn't work, but it seems to now.
2020-03-25 09:52:03 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[PTRTOINT]], [[C]](s32)
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[PTRTOINT]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[PTRTOINT]], [[C2]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[PTRTOINT]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
AMDGPU/GlobalISel: Add a testcase for G_UNMERGE_VALUES legalization I had a note that this doesn't work, but it seems to now.
2020-03-25 09:52:03 +08:00
%0:_(p3) = COPY $vgpr0
%1:_(s8), %2:_(s8), %3:_(s8), %4:_(s8) = G_UNMERGE_VALUES %0
%5:_(s32) = G_ANYEXT %1
%6:_(s32) = G_ANYEXT %2
%7:_(s32) = G_ANYEXT %3
%8:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
$vgpr2 = COPY %7
$vgpr3 = COPY %8
...
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
---
name: test_unmerge_s16_s64
body: |
bb.0:
liveins: $vgpr0_vgpr1
; CHECK-LABEL: name: test_unmerge_s16_s64
; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
GlobalISel: Reimplement widenScalar for G_UNMERGE_VALUES results Only use shifts if the requested type exactly matches the source type, and create sub-unmerges otherwise.
2020-01-21 22:02:42 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV]], [[C]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV1]], [[C]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s16), %2:_(s16), %3:_(s16), %4:_(s16) = G_UNMERGE_VALUES %0
%5:_(s32) = G_ANYEXT %1
%6:_(s32) = G_ANYEXT %2
%7:_(s32) = G_ANYEXT %3
%8:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
$vgpr2 = COPY %7
$vgpr3 = COPY %8
...
---
name: test_unmerge_s1_s3
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s1_s3
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
[GlobalISel] support widen unmerge if WideTy > SrcTy Summary: Widening G_UNMERGE_VALUES to a type which is larger than the original source type is the same as widening it to the same type as the source type: in both cases, G_UNMERGE_VALUES has to be replaced with bit arithmetic which. Although the arithmetic itself is independent of whether the source type is smaller or equal to the widen type, widening the source type to the widen type should result in less artifacts being emitted, since this is the type that the user explicitly requested. Reviewers: arsenm, dsanders, aemerson, aditya_nandakumar Reviewed By: arsenm, dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, hiraditya, volkan, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D76494
2020-03-20 21:46:01 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
; CHECK: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C3]](s32)
; CHECK: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
; CHECK: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C4]](s32)
; CHECK: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
; CHECK: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C5]](s32)
; CHECK: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
; CHECK: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C6]](s32)
; CHECK: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C7]](s32)
; CHECK: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 9
; CHECK: [[LSHR8:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C8]](s32)
; CHECK: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 10
; CHECK: [[LSHR9:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C9]](s32)
; CHECK: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 11
; CHECK: [[LSHR10:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C10]](s32)
; CHECK: [[C11:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
; CHECK: [[LSHR11:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C11]](s32)
; CHECK: [[C12:%[0-9]+]]:_(s32) = G_CONSTANT i32 13
; CHECK: [[LSHR12:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C12]](s32)
; CHECK: [[C13:%[0-9]+]]:_(s32) = G_CONSTANT i32 14
; CHECK: [[LSHR13:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C13]](s32)
; CHECK: [[C14:%[0-9]+]]:_(s32) = G_CONSTANT i32 15
; CHECK: [[LSHR14:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C14]](s32)
; CHECK: [[C15:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR15:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C15]](s32)
; CHECK: [[C16:%[0-9]+]]:_(s32) = G_CONSTANT i32 17
; CHECK: [[LSHR16:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C16]](s32)
; CHECK: [[C17:%[0-9]+]]:_(s32) = G_CONSTANT i32 18
; CHECK: [[LSHR17:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C17]](s32)
; CHECK: [[C18:%[0-9]+]]:_(s32) = G_CONSTANT i32 19
; CHECK: [[LSHR18:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C18]](s32)
; CHECK: [[C19:%[0-9]+]]:_(s32) = G_CONSTANT i32 20
; CHECK: [[LSHR19:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C19]](s32)
; CHECK: [[C20:%[0-9]+]]:_(s32) = G_CONSTANT i32 21
; CHECK: [[LSHR20:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C20]](s32)
; CHECK: [[C21:%[0-9]+]]:_(s32) = G_CONSTANT i32 22
; CHECK: [[LSHR21:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C21]](s32)
; CHECK: [[C22:%[0-9]+]]:_(s32) = G_CONSTANT i32 23
; CHECK: [[LSHR22:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C22]](s32)
; CHECK: [[C23:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR23:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C23]](s32)
; CHECK: [[C24:%[0-9]+]]:_(s32) = G_CONSTANT i32 25
; CHECK: [[LSHR24:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C24]](s32)
; CHECK: [[C25:%[0-9]+]]:_(s32) = G_CONSTANT i32 26
; CHECK: [[LSHR25:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C25]](s32)
; CHECK: [[C26:%[0-9]+]]:_(s32) = G_CONSTANT i32 27
; CHECK: [[LSHR26:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C26]](s32)
; CHECK: [[C27:%[0-9]+]]:_(s32) = G_CONSTANT i32 28
; CHECK: [[LSHR27:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C27]](s32)
; CHECK: [[C28:%[0-9]+]]:_(s32) = G_CONSTANT i32 29
; CHECK: [[LSHR28:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C28]](s32)
; CHECK: [[C29:%[0-9]+]]:_(s32) = G_CONSTANT i32 30
; CHECK: [[LSHR29:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C29]](s32)
; CHECK: [[C30:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
; CHECK: [[LSHR30:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C30]](s32)
[GlobalISel] support widen unmerge if WideTy > SrcTy Summary: Widening G_UNMERGE_VALUES to a type which is larger than the original source type is the same as widening it to the same type as the source type: in both cases, G_UNMERGE_VALUES has to be replaced with bit arithmetic which. Although the arithmetic itself is independent of whether the source type is smaller or equal to the widen type, widening the source type to the widen type should result in less artifacts being emitted, since this is the type that the user explicitly requested. Reviewers: arsenm, dsanders, aemerson, aditya_nandakumar Reviewed By: arsenm, dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, hiraditya, volkan, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D76494
2020-03-20 21:46:01 +08:00
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s3) = G_TRUNC %0
%2:_(s1), %3:_(s1), %4:_(s1) = G_UNMERGE_VALUES %1
%5:_(s32) = G_ANYEXT %2
%6:_(s32) = G_ANYEXT %3
%7:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
$vgpr2 = COPY %7
...
---
name: test_unmerge_s1_s8
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s1_s8
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
[GlobalISel] support widen unmerge if WideTy > SrcTy Summary: Widening G_UNMERGE_VALUES to a type which is larger than the original source type is the same as widening it to the same type as the source type: in both cases, G_UNMERGE_VALUES has to be replaced with bit arithmetic which. Although the arithmetic itself is independent of whether the source type is smaller or equal to the widen type, widening the source type to the widen type should result in less artifacts being emitted, since this is the type that the user explicitly requested. Reviewers: arsenm, dsanders, aemerson, aditya_nandakumar Reviewed By: arsenm, dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, hiraditya, volkan, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D76494
2020-03-20 21:46:01 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 2
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 3
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
; CHECK: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
; CHECK: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C3]](s32)
; CHECK: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 5
; CHECK: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C4]](s32)
; CHECK: [[C5:%[0-9]+]]:_(s32) = G_CONSTANT i32 6
; CHECK: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C5]](s32)
; CHECK: [[C6:%[0-9]+]]:_(s32) = G_CONSTANT i32 7
; CHECK: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C6]](s32)
; CHECK: [[C7:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C7]](s32)
; CHECK: [[C8:%[0-9]+]]:_(s32) = G_CONSTANT i32 9
; CHECK: [[LSHR8:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C8]](s32)
; CHECK: [[C9:%[0-9]+]]:_(s32) = G_CONSTANT i32 10
; CHECK: [[LSHR9:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C9]](s32)
; CHECK: [[C10:%[0-9]+]]:_(s32) = G_CONSTANT i32 11
; CHECK: [[LSHR10:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C10]](s32)
; CHECK: [[C11:%[0-9]+]]:_(s32) = G_CONSTANT i32 12
; CHECK: [[LSHR11:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C11]](s32)
; CHECK: [[C12:%[0-9]+]]:_(s32) = G_CONSTANT i32 13
; CHECK: [[LSHR12:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C12]](s32)
; CHECK: [[C13:%[0-9]+]]:_(s32) = G_CONSTANT i32 14
; CHECK: [[LSHR13:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C13]](s32)
; CHECK: [[C14:%[0-9]+]]:_(s32) = G_CONSTANT i32 15
; CHECK: [[LSHR14:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C14]](s32)
; CHECK: [[C15:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR15:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C15]](s32)
; CHECK: [[C16:%[0-9]+]]:_(s32) = G_CONSTANT i32 17
; CHECK: [[LSHR16:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C16]](s32)
; CHECK: [[C17:%[0-9]+]]:_(s32) = G_CONSTANT i32 18
; CHECK: [[LSHR17:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C17]](s32)
; CHECK: [[C18:%[0-9]+]]:_(s32) = G_CONSTANT i32 19
; CHECK: [[LSHR18:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C18]](s32)
; CHECK: [[C19:%[0-9]+]]:_(s32) = G_CONSTANT i32 20
; CHECK: [[LSHR19:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C19]](s32)
; CHECK: [[C20:%[0-9]+]]:_(s32) = G_CONSTANT i32 21
; CHECK: [[LSHR20:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C20]](s32)
; CHECK: [[C21:%[0-9]+]]:_(s32) = G_CONSTANT i32 22
; CHECK: [[LSHR21:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C21]](s32)
; CHECK: [[C22:%[0-9]+]]:_(s32) = G_CONSTANT i32 23
; CHECK: [[LSHR22:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C22]](s32)
; CHECK: [[C23:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR23:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C23]](s32)
; CHECK: [[C24:%[0-9]+]]:_(s32) = G_CONSTANT i32 25
; CHECK: [[LSHR24:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C24]](s32)
; CHECK: [[C25:%[0-9]+]]:_(s32) = G_CONSTANT i32 26
; CHECK: [[LSHR25:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C25]](s32)
; CHECK: [[C26:%[0-9]+]]:_(s32) = G_CONSTANT i32 27
; CHECK: [[LSHR26:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C26]](s32)
; CHECK: [[C27:%[0-9]+]]:_(s32) = G_CONSTANT i32 28
; CHECK: [[LSHR27:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C27]](s32)
; CHECK: [[C28:%[0-9]+]]:_(s32) = G_CONSTANT i32 29
; CHECK: [[LSHR28:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C28]](s32)
; CHECK: [[C29:%[0-9]+]]:_(s32) = G_CONSTANT i32 30
; CHECK: [[LSHR29:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C29]](s32)
; CHECK: [[C30:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
; CHECK: [[LSHR30:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C30]](s32)
[GlobalISel] support widen unmerge if WideTy > SrcTy Summary: Widening G_UNMERGE_VALUES to a type which is larger than the original source type is the same as widening it to the same type as the source type: in both cases, G_UNMERGE_VALUES has to be replaced with bit arithmetic which. Although the arithmetic itself is independent of whether the source type is smaller or equal to the widen type, widening the source type to the widen type should result in less artifacts being emitted, since this is the type that the user explicitly requested. Reviewers: arsenm, dsanders, aemerson, aditya_nandakumar Reviewed By: arsenm, dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, hiraditya, volkan, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D76494
2020-03-20 21:46:01 +08:00
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[LSHR4]](s32)
; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[LSHR5]](s32)
; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR6]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
; CHECK: $vgpr4 = COPY [[COPY5]](s32)
; CHECK: $vgpr5 = COPY [[COPY6]](s32)
; CHECK: $vgpr6 = COPY [[COPY7]](s32)
; CHECK: $vgpr7 = COPY [[COPY8]](s32)
GlobalISel: Implement widenScalar for G_UNMERGE_VALUES For the scalar case only. Also move the similar G_MERGE_VALUES handling to a separate function and cleanup to make them look more similar. llvm-svn: 352979
2019-02-03 08:07:33 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s8) = G_TRUNC %0
%2:_(s1), %3:_(s1), %4:_(s1), %5:_(s1), %6:_(s1), %7:_(s1), %8:_(s1), %9:_(s1) = G_UNMERGE_VALUES %1
%10:_(s32) = G_ANYEXT %2
%11:_(s32) = G_ANYEXT %3
%12:_(s32) = G_ANYEXT %4
%13:_(s32) = G_ANYEXT %5
%14:_(s32) = G_ANYEXT %6
%15:_(s32) = G_ANYEXT %7
%16:_(s32) = G_ANYEXT %8
%17:_(s32) = G_ANYEXT %9
$vgpr0 = COPY %10
$vgpr1 = COPY %11
$vgpr2 = COPY %12
$vgpr3 = COPY %13
$vgpr4 = COPY %14
$vgpr5 = COPY %15
$vgpr6 = COPY %16
$vgpr7 = COPY %17
...
AMDGPU/GlobalISel: Make some large merges legal We allow up to 1024-bit registers, so we should support merges all the way to the maximum.
2020-03-16 22:42:15 +08:00
---
name: test_unmerge_s128_v2s128
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
; CHECK-LABEL: name: test_unmerge_s128_v2s128
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s128>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
; CHECK: [[UV:%[0-9]+]]:_(s128), [[UV1:%[0-9]+]]:_(s128) = G_UNMERGE_VALUES [[COPY]](<2 x s128>)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[UV]](s128)
; CHECK: $vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[UV1]](s128)
%0:_(<2 x s128>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
%1:_(s128), %2:_(s128) = G_UNMERGE_VALUES %0
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
$vgpr4_vgpr5_vgpr6_vgpr7 = COPY %2
...
---
name: test_unmerge_s128_s256
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
; CHECK-LABEL: name: test_unmerge_s128_s256
; CHECK: [[COPY:%[0-9]+]]:_(s256) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
; CHECK: [[UV:%[0-9]+]]:_(s128), [[UV1:%[0-9]+]]:_(s128) = G_UNMERGE_VALUES [[COPY]](s256)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[UV]](s128)
; CHECK: $vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[UV1]](s128)
%0:_(s256) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
%1:_(s128), %2:_(s128) = G_UNMERGE_VALUES %0
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
$vgpr4_vgpr5_vgpr6_vgpr7 = COPY %2
...
---
name: test_unmerge_s256_s512
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; CHECK-LABEL: name: test_unmerge_s256_s512
; CHECK: [[COPY:%[0-9]+]]:_(s512) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; CHECK: [[UV:%[0-9]+]]:_(s256), [[UV1:%[0-9]+]]:_(s256) = G_UNMERGE_VALUES [[COPY]](s512)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[UV]](s256)
; CHECK: $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[UV1]](s256)
%0:_(s512) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
%1:_(s256), %2:_(s256) = G_UNMERGE_VALUES %0
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY %1
$vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %2
...
---
name: test_unmerge_s256_v2s256
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; CHECK-LABEL: name: test_unmerge_s256_v2s256
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s256>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
; CHECK: [[UV:%[0-9]+]]:_(s256), [[UV1:%[0-9]+]]:_(s256) = G_UNMERGE_VALUES [[COPY]](<2 x s256>)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY [[UV]](s256)
; CHECK: $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[UV1]](s256)
%0:_(<2 x s256>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
%1:_(s256), %2:_(s256) = G_UNMERGE_VALUES %0
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = COPY %1
$vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %2
...
---
name: test_unmerge_s512_s1024
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
; CHECK-LABEL: name: test_unmerge_s512_s1024
; CHECK: [[COPY:%[0-9]+]]:_(s1024) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
; CHECK: [[UV:%[0-9]+]]:_(s512), [[UV1:%[0-9]+]]:_(s512) = G_UNMERGE_VALUES [[COPY]](s1024)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[UV]](s512)
; CHECK: $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY [[UV1]](s512)
%0:_(s1024) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
%1:_(s512), %2:_(s512) = G_UNMERGE_VALUES %0
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %1
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY %2
...
---
name: test_unmerge_s512_v2s512
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
; CHECK-LABEL: name: test_unmerge_s512_v2s512
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s512>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
; CHECK: [[UV:%[0-9]+]]:_(s512), [[UV1:%[0-9]+]]:_(s512) = G_UNMERGE_VALUES [[COPY]](<2 x s512>)
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY [[UV]](s512)
; CHECK: $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY [[UV1]](s512)
%0:_(<2 x s512>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31
%1:_(s512), %2:_(s512) = G_UNMERGE_VALUES %0
$vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = COPY %1
$vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = COPY %2
...
GlobalISel: Combine G_UNMERGE_VALUES with G_TRUNC G_BITCAST can be lowered with a pair of G_UNMERGE_VALUES and G_MERGE_VALUES with different types, but G_UNMERGE_VALUES of a vector can also be implemented with a bitcast to a scalar, which introduces the possibility for infinite loops. Try to eliminate an illegal source register type in the artifact combiner to avoid this from happening. Avoids infinite looping in the legalizer in a future patch which allows lowering G_UNMERGE_VALUES of a vector source with a G_BITCAST.
2020-04-05 03:07:28 +08:00
---
name: test_unmerge_v2s1
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_v2s1
; CHECK: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
; CHECK: [[TRUNC:%[0-9]+]]:_(s1) = G_TRUNC [[UV]](s32)
; CHECK: [[TRUNC1:%[0-9]+]]:_(s1) = G_TRUNC [[UV1]](s32)
; CHECK: S_NOP 0, implicit [[TRUNC]](s1)
; CHECK: S_NOP 0, implicit [[TRUNC1]](s1)
%0:_(<2 x s1>) = G_IMPLICIT_DEF
%1:_(s1), %2:_(s1) = G_UNMERGE_VALUES %0
S_NOP 0, implicit %1
S_NOP 0, implicit %2
...
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
name: test_unmerge_s8_v4s8
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s8_v4s8
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
; CHECK: $vgpr3 = COPY [[COPY4]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(<4 x s8>) = G_BITCAST %0
%2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8) = G_UNMERGE_VALUES %1
%6:_(s32) = G_ANYEXT %2
%7:_(s32) = G_ANYEXT %3
%8:_(s32) = G_ANYEXT %4
%9:_(s32) = G_ANYEXT %5
$vgpr0 = COPY %6
$vgpr1 = COPY %7
$vgpr2 = COPY %8
$vgpr3 = COPY %9
...
---
name: test_unmerge_s8_v3s8
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s8_v3s8
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
; CHECK: $vgpr2 = COPY [[COPY3]](s32)
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s24) = G_TRUNC %0
%2:_(<3 x s8>) = G_BITCAST %1
%3:_(s8), %4:_(s8), %5:_(s8) = G_UNMERGE_VALUES %2
%6:_(s32) = G_ANYEXT %3
%7:_(s32) = G_ANYEXT %4
%8:_(s32) = G_ANYEXT %5
$vgpr0 = COPY %6
$vgpr1 = COPY %7
$vgpr2 = COPY %8
...
---
name: test_unmerge_s8_v2s8
body: |
bb.0:
liveins: $vgpr0
; CHECK-LABEL: name: test_unmerge_s8_v2s8
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
; CHECK: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
[GlobalISel] Combine scalar unmerge(trunc) Summary: Combine unmerge(trunc) to enable other merge combines. Without this combine, the scalar unmerge(trunc(merge)) pattern cannot be combined and easily lead to hard-to-legalize merge/unmerge artifacts. Reviewed By: arsenm Tags: #llvm Differential Revision: https://reviews.llvm.org/D79567
2020-05-07 18:27:57 +08:00
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: $vgpr0 = COPY [[COPY1]](s32)
; CHECK: $vgpr1 = COPY [[COPY2]](s32)
GlobalISel: Handle more cases in lowerUnmergeValues Handle scalar sources, as well as vectors.
2020-03-30 06:04:53 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
%2:_(<2 x s8>) = G_BITCAST %1
%3:_(s8), %4:_(s8) = G_UNMERGE_VALUES %2
%5:_(s32) = G_ANYEXT %3
%6:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
$vgpr1 = COPY %6
...
AMDGPU/GlobalISel: Ensure subreg is valid when selecting G_UNMERGE_VALUES Fixes verifier error with SGPR unmerges with 96-bit result types.
2020-08-04 21:19:43 +08:00
---
name: test_unmerge_v3s32_v12s32
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
; CHECK-LABEL: name: test_unmerge_v3s32_v12s32
; CHECK: [[COPY:%[0-9]+]]:_(<6 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5
; CHECK: [[COPY1:%[0-9]+]]:_(<6 x s32>) = COPY $vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
; CHECK: [[UV:%[0-9]+]]:_(<3 x s32>), [[UV1:%[0-9]+]]:_(<3 x s32>) = G_UNMERGE_VALUES [[COPY]](<6 x s32>)
; CHECK: [[UV2:%[0-9]+]]:_(<3 x s32>), [[UV3:%[0-9]+]]:_(<3 x s32>) = G_UNMERGE_VALUES [[COPY1]](<6 x s32>)
; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[UV]](<3 x s32>)
; CHECK: $vgpr3_vgpr4_vgpr5 = COPY [[UV1]](<3 x s32>)
; CHECK: $vgpr6_vgpr7_vgpr8 = COPY [[UV2]](<3 x s32>)
; CHECK: $vgpr9_vgpr10_vgpr11 = COPY [[UV3]](<3 x s32>)
%0:_(<6 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5
%1:_(<6 x s32>) = COPY $vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11
%2:_(<12 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(<3 x s32>), %4:_(<3 x s32>), %5:_(<3 x s32>), %6:_(<3 x s32>) = G_UNMERGE_VALUES %2
$vgpr0_vgpr1_vgpr2 = COPY %3
$vgpr3_vgpr4_vgpr5 = COPY %4
$vgpr6_vgpr7_vgpr8 = COPY %5
$vgpr9_vgpr10_vgpr11 = COPY %6
...