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/artifact-combiner-unmerge-v...

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

539 lines
29 KiB
Plaintext
Raw Normal View History

GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -O0 -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 -o - %s | FileCheck %s
---
name: test_unmerge_values_s1_trunc_v2s1_of_build_vector_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s1_trunc_v2s1_of_build_vector_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[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: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV]](s32), [[UV2]]
; CHECK: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV1]](s32), [[UV3]]
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ANYEXT]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; CHECK: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 1
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[ANYEXT1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; CHECK: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 1
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG]](s32), [[SEXT_INREG1]](s32)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr0_vgpr1
%2:_(s32), %3:_(s32) = G_UNMERGE_VALUES %0(<2 x s32>)
%4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %1(<2 x s32>)
%6:_(s1) = G_ICMP intpred(ne), %2(s32), %4
%7:_(s1) = G_ICMP intpred(ne), %3(s32), %5
%8:_(s32) = G_ANYEXT %6(s1)
%9:_(s32) = G_ANYEXT %7(s1)
%10:_(<2 x s32>) = G_BUILD_VECTOR %8, %9
%11:_(<2 x s1>) = G_TRUNC %10(<2 x s32>)
%12:_(s1), %13:_(s1) = G_UNMERGE_VALUES %11
%14:_(s32) = G_SEXT %12
%15:_(s32) = G_SEXT %13
%16:_(<2 x s32>) = G_BUILD_VECTOR %14, %15
$vgpr0_vgpr1 = COPY %16
...
# Requires looking thorugh extra copies between the build_vector,
# trunc and unmerge.
---
name: test_unmerge_values_s1_trunc_v2s1_of_build_vector_v2s32_extra_copies
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s1_trunc_v2s1_of_build_vector_v2s32_extra_copies
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[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: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV]](s32), [[UV2]]
; CHECK: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV1]](s32), [[UV3]]
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ANYEXT]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; CHECK: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 1
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[ANYEXT1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; CHECK: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 1
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG]](s32), [[SEXT_INREG1]](s32)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr0_vgpr1
%2:_(s32), %3:_(s32) = G_UNMERGE_VALUES %0(<2 x s32>)
%4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %1(<2 x s32>)
%6:_(s1) = G_ICMP intpred(ne), %2(s32), %4
%7:_(s1) = G_ICMP intpred(ne), %3(s32), %5
%8:_(s32) = G_ANYEXT %6(s1)
%9:_(s32) = G_ANYEXT %7(s1)
%10:_(<2 x s32>) = G_BUILD_VECTOR %8, %9
%11:_(<2 x s32>) = COPY %10
%12:_(<2 x s1>) = G_TRUNC %11(<2 x s32>)
%13:_(<2 x s1>) = COPY %12
%14:_(s1), %15:_(s1) = G_UNMERGE_VALUES %13
%16:_(s32) = G_SEXT %14
%17:_(s32) = G_SEXT %15
%18:_(<2 x s32>) = G_BUILD_VECTOR %16, %17
$vgpr0_vgpr1 = COPY %18
...
---
name: test_unmerge_values_s32_sext_v2s32_of_build_vector_v2s16
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s32_sext_v2s32_of_build_vector_v2s16
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[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: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV]](s32), [[UV2]]
; CHECK: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV1]](s32), [[UV3]]
; CHECK: [[ANYEXT:%[0-9]+]]:_(s16) = G_ANYEXT [[ICMP]](s1)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s16) = G_ANYEXT [[ICMP1]](s1)
; CHECK: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[ANYEXT]](s16)
; CHECK: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[ANYEXT1]](s16)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT]](s32), [[SEXT1]](s32)
; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr0_vgpr1
%2:_(s32), %3:_(s32) = G_UNMERGE_VALUES %0(<2 x s32>)
%4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %1(<2 x s32>)
%6:_(s1) = G_ICMP intpred(ne), %2(s32), %4
%7:_(s1) = G_ICMP intpred(ne), %3(s32), %5
%8:_(s16) = G_ANYEXT %6
%9:_(s16) = G_ANYEXT %7
%10:_(<2 x s16>) = G_BUILD_VECTOR %8, %9
%11:_(<2 x s32>) = G_SEXT %10
%12:_(s32), %13:_(s32) = G_UNMERGE_VALUES %11
%14:_(<2 x s32>) = G_BUILD_VECTOR %12, %13
$vgpr0_vgpr1 = COPY %14
...
---
name: test_unmerge_values_s32_zext_v2s32_of_build_vector_v2s16
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s32_zext_v2s32_of_build_vector_v2s16
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[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: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV]](s32), [[UV2]]
; CHECK: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV1]](s32), [[UV3]]
; CHECK: [[ANYEXT:%[0-9]+]]:_(s16) = G_ANYEXT [[ICMP]](s1)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s16) = G_ANYEXT [[ICMP1]](s1)
; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ANYEXT]](s16)
; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[ANYEXT1]](s16)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[ZEXT]](s32), [[ZEXT1]](s32)
; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr0_vgpr1
%2:_(s32), %3:_(s32) = G_UNMERGE_VALUES %0(<2 x s32>)
%4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %1(<2 x s32>)
%6:_(s1) = G_ICMP intpred(ne), %2(s32), %4
%7:_(s1) = G_ICMP intpred(ne), %3(s32), %5
%8:_(s16) = G_ANYEXT %6(s1)
%9:_(s16) = G_ANYEXT %7(s1)
%10:_(<2 x s16>) = G_BUILD_VECTOR %8, %9
%11:_(<2 x s32>) = G_ZEXT %10
%12:_(s32), %13:_(s32) = G_UNMERGE_VALUES %11
%14:_(<2 x s32>) = G_BUILD_VECTOR %12(s32), %13(s32)
$vgpr0_vgpr1 = COPY %14(<2 x s32>)
...
---
name: test_unmerge_values_s32_anyext_v2s32_of_build_vector_v2s16
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s32_anyext_v2s32_of_build_vector_v2s16
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[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: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV]](s32), [[UV2]]
; CHECK: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV1]](s32), [[UV3]]
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP]](s1)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ICMP1]](s1)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32)
; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr0_vgpr1
%2:_(s32), %3:_(s32) = G_UNMERGE_VALUES %0(<2 x s32>)
%4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %1(<2 x s32>)
%6:_(s1) = G_ICMP intpred(ne), %2(s32), %4
%7:_(s1) = G_ICMP intpred(ne), %3(s32), %5
%8:_(s16) = G_ANYEXT %6(s1)
%9:_(s16) = G_ANYEXT %7(s1)
%10:_(<2 x s16>) = G_BUILD_VECTOR %8, %9
%11:_(<2 x s32>) = G_ANYEXT %10
%12:_(s32), %13:_(s32) = G_UNMERGE_VALUES %11
%14:_(<2 x s32>) = G_BUILD_VECTOR %12, %13
$vgpr0_vgpr1 = COPY %14
...
---
name: test_unmerge_values_v2s16_zext_v4s32_of_build_vector_v4s16
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_v2s16_zext_v4s32_of_build_vector_v4s16
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[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: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV]](s32), [[UV2]]
; CHECK: [[ICMP1:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[UV1]](s32), [[UV3]]
; CHECK: [[ANYEXT:%[0-9]+]]:_(s16) = G_ANYEXT [[ICMP]](s1)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s16) = G_ANYEXT [[ICMP1]](s1)
; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[ANYEXT]](s16)
; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[ANYEXT1]](s16)
; CHECK: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[ANYEXT]](s16)
; CHECK: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[ANYEXT1]](s16)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[ZEXT]](s32), [[ZEXT1]](s32), [[ZEXT2]](s32), [[ZEXT3]](s32)
; CHECK: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>), [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
; CHECK: S_ENDPGM 0, implicit [[UV4]](<2 x s16>), implicit [[UV5]](<2 x s16>), implicit [[UV6]](<2 x s16>), implicit [[UV7]](<2 x s16>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr0_vgpr1
%2:_(s32), %3:_(s32) = G_UNMERGE_VALUES %0(<2 x s32>)
%4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %1(<2 x s32>)
%6:_(s1) = G_ICMP intpred(ne), %2(s32), %4
%7:_(s1) = G_ICMP intpred(ne), %3(s32), %5
%8:_(s16) = G_ANYEXT %6
%9:_(s16) = G_ANYEXT %7
%10:_(<4 x s16>) = G_BUILD_VECTOR %8, %9, %8, %9
%11:_(<4 x s32>) = G_ZEXT %10
%12:_(<2 x s16>), %13:_(<2 x s16>), %14:_(<2 x s16>), %15:_(<2 x s16>) = G_UNMERGE_VALUES %11
S_ENDPGM 0, implicit %12, implicit %13, implicit %14, implicit %15
...
---
name: test_unmerge_values_s1_trunc_v4s1_of_concat_vectors_v4s32_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s1_trunc_v4s1_of_concat_vectors_v4s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s32>) = G_CONCAT_VECTORS [[COPY]](<2 x s32>), [[COPY1]](<2 x s32>)
; CHECK: [[TRUNC:%[0-9]+]]:_(<4 x s1>) = G_TRUNC [[CONCAT_VECTORS]](<4 x s32>)
; CHECK: [[UV:%[0-9]+]]:_(s1), [[UV1:%[0-9]+]]:_(s1), [[UV2:%[0-9]+]]:_(s1), [[UV3:%[0-9]+]]:_(s1) = G_UNMERGE_VALUES [[TRUNC]](<4 x s1>)
; CHECK: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[UV]](s1)
; CHECK: [[SEXT1:%[0-9]+]]:_(s32) = G_SEXT [[UV1]](s1)
; CHECK: [[SEXT2:%[0-9]+]]:_(s32) = G_SEXT [[UV2]](s1)
; CHECK: [[SEXT3:%[0-9]+]]:_(s32) = G_SEXT [[UV3]](s1)
; CHECK: $vgpr0 = COPY [[SEXT]](s32)
; CHECK: $vgpr1 = COPY [[SEXT1]](s32)
; CHECK: $vgpr2 = COPY [[SEXT2]](s32)
; CHECK: $vgpr3 = COPY [[SEXT3]](s32)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<4 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s1>) = G_TRUNC %2
%4:_(s1), %5:_(s1), %6:_(s1), %7:_(s1) = G_UNMERGE_VALUES %3
%8:_(s32) = G_SEXT %4
%9:_(s32) = G_SEXT %5
%10:_(s32) = G_SEXT %6
%11:_(s32) = G_SEXT %7
$vgpr0 = COPY %8
$vgpr1 = COPY %9
$vgpr2 = COPY %10
$vgpr3 = COPY %11
...
---
name: test_unmerge_values_s16_of_concat_vectors_v2s16_v2s16
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s16_of_concat_vectors_v2s16_v2s16
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
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: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; CHECK: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; CHECK: S_ENDPGM 0, implicit [[TRUNC]](s16), implicit [[TRUNC1]](s16), implicit [[TRUNC2]](s16), implicit [[TRUNC3]](s16)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s16>) = COPY $vgpr0
%1:_(<2 x s16>) = COPY $vgpr1
%2:_(<4 x s16>) = G_CONCAT_VECTORS %0, %1
%3:_(s16), %4:_(s16), %5:_(s16), %6:_(s16) = G_UNMERGE_VALUES %2
S_ENDPGM 0, implicit %3, implicit %4, implicit %5, implicit %6
...
---
name: test_unmerge_values_s32_of_concat_vectors_v2s32_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s32_of_concat_vectors_v2s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr1_vgpr2
GlobalISel: Fold out unmerge to scalars from concat_vector Removes illegal intermediate vectors if an operation was lowering to concat_vectors, and the next operation is scalarized. llvm-svn: 367081
2019-07-26 10:22:23 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: S_ENDPGM 0, implicit [[UV]](s32), implicit [[UV1]](s32), implicit [[UV2]](s32), implicit [[UV3]](s32)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr1_vgpr2
%2:_(<4 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(s32), %4:_(s32), %5:_(s32), %6:_(s32) = G_UNMERGE_VALUES %2
S_ENDPGM 0, implicit %3, implicit %4, implicit %5, implicit %6
...
---
name: test_unmerge_values_s32_of_concat_vectors_v2s64_v2s64
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s32_of_concat_vectors_v2s64_v2s64
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
GlobalISel: Fold out unmerge to scalars from concat_vector Removes illegal intermediate vectors if an operation was lowering to concat_vectors, and the next operation is scalarized. llvm-svn: 367081
2019-07-26 10:22:23 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; CHECK: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: S_ENDPGM 0, implicit [[UV]](s32), implicit [[UV1]](s32), implicit [[UV2]](s32), implicit [[UV3]](s32), implicit [[UV4]](s32), implicit [[UV5]](s32), implicit [[UV6]](s32), implicit [[UV7]](s32)
%0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
%2:_(<4 x s64>) = G_CONCAT_VECTORS %0, %1
%3:_(s32), %4:_(s32), %5:_(s32), %6:_(s32), %7:_(s32), %8:_(s32), %9:_(s32), %10:_(s32) = G_UNMERGE_VALUES %2
S_ENDPGM 0, implicit %3, implicit %4, implicit %5, implicit %6, implicit %7, implicit %8, implicit %9, implicit %10
...
---
name: test_unmerge_values_s32_of_trunc_concat_vectors_v2s64_v2s64
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s32_of_trunc_concat_vectors_v2s64_v2s64
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
AMDGPU/GlobalISel: Refine G_TRUNC legality rules Scalarize most truncates. Avoid touching cases that could end up in unresolvable infinite loops.
2020-02-16 10:59:30 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; CHECK: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[UV]](s64)
; CHECK: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[UV1]](s64)
; CHECK: [[TRUNC2:%[0-9]+]]:_(s32) = G_TRUNC [[UV2]](s64)
; CHECK: [[TRUNC3:%[0-9]+]]:_(s32) = G_TRUNC [[UV3]](s64)
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[TRUNC]](s32), [[TRUNC1]](s32), [[TRUNC2]](s32), [[TRUNC3]](s32)
; CHECK: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
; CHECK: S_ENDPGM 0, implicit [[UV4]](s32), implicit [[UV5]](s32), implicit [[UV6]](s32), implicit [[UV7]](s32)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
%2:_(<4 x s64>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s32>) = G_TRUNC %2
%4:_(s32), %5:_(s32), %6:_(s32), %7:_(s32) = G_UNMERGE_VALUES %3
S_ENDPGM 0, implicit %4, implicit %5, implicit %6, implicit %7
...
---
name: test_unmerge_values_s64_of_sext_concat_vectors_v2s32_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s64_of_sext_concat_vectors_v2s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
GlobalISel: Fold out unmerge to scalars from concat_vector Removes illegal intermediate vectors if an operation was lowering to concat_vectors, and the next operation is scalarized. llvm-svn: 367081
2019-07-26 10:22:23 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[UV]](s32)
; CHECK: [[SEXT1:%[0-9]+]]:_(s64) = G_SEXT [[UV1]](s32)
; CHECK: [[SEXT2:%[0-9]+]]:_(s64) = G_SEXT [[UV2]](s32)
; CHECK: [[SEXT3:%[0-9]+]]:_(s64) = G_SEXT [[UV3]](s32)
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[SEXT]](s64), [[SEXT1]](s64), [[SEXT2]](s64), [[SEXT3]](s64)
; CHECK: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64), [[UV6:%[0-9]+]]:_(s64), [[UV7:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s64>)
; CHECK: S_ENDPGM 0, implicit [[UV4]](s64), implicit [[UV5]](s64), implicit [[UV6]](s64), implicit [[UV7]](s64)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<4 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s64>) = G_SEXT %2
%4:_(s64), %5:_(s64), %6:_(s64), %7:_(s64) = G_UNMERGE_VALUES %3
S_ENDPGM 0, implicit %4, implicit %5, implicit %6, implicit %7
...
---
name: test_unmerge_values_s64_of_zext_concat_vectors_v2s32_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s64_of_zext_concat_vectors_v2s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
GlobalISel: Fold out unmerge to scalars from concat_vector Removes illegal intermediate vectors if an operation was lowering to concat_vectors, and the next operation is scalarized. llvm-svn: 367081
2019-07-26 10:22:23 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[UV]](s32)
; CHECK: [[ZEXT1:%[0-9]+]]:_(s64) = G_ZEXT [[UV1]](s32)
; CHECK: [[ZEXT2:%[0-9]+]]:_(s64) = G_ZEXT [[UV2]](s32)
; CHECK: [[ZEXT3:%[0-9]+]]:_(s64) = G_ZEXT [[UV3]](s32)
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[ZEXT]](s64), [[ZEXT1]](s64), [[ZEXT2]](s64), [[ZEXT3]](s64)
; CHECK: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64), [[UV6:%[0-9]+]]:_(s64), [[UV7:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s64>)
; CHECK: S_ENDPGM 0, implicit [[UV4]](s64), implicit [[UV5]](s64), implicit [[UV6]](s64), implicit [[UV7]](s64)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<4 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s64>) = G_ZEXT %2
%4:_(s64), %5:_(s64), %6:_(s64), %7:_(s64) = G_UNMERGE_VALUES %3
S_ENDPGM 0, implicit %4, implicit %5, implicit %6, implicit %7
...
---
name: test_unmerge_values_s64_of_anyext_concat_vectors_v2s32_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s64_of_anyext_concat_vectors_v2s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
GlobalISel: Fold out unmerge to scalars from concat_vector Removes illegal intermediate vectors if an operation was lowering to concat_vectors, and the next operation is scalarized. llvm-svn: 367081
2019-07-26 10:22:23 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[UV]](s32)
; CHECK: [[ANYEXT1:%[0-9]+]]:_(s64) = G_ANYEXT [[UV1]](s32)
; CHECK: [[ANYEXT2:%[0-9]+]]:_(s64) = G_ANYEXT [[UV2]](s32)
; CHECK: [[ANYEXT3:%[0-9]+]]:_(s64) = G_ANYEXT [[UV3]](s32)
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s64>) = G_BUILD_VECTOR [[ANYEXT]](s64), [[ANYEXT1]](s64), [[ANYEXT2]](s64), [[ANYEXT3]](s64)
; CHECK: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64), [[UV6:%[0-9]+]]:_(s64), [[UV7:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s64>)
; CHECK: S_ENDPGM 0, implicit [[UV4]](s64), implicit [[UV5]](s64), implicit [[UV6]](s64), implicit [[UV7]](s64)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<4 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s64>) = G_ANYEXT %2
%4:_(s64), %5:_(s64), %6:_(s64), %7:_(s64) = G_UNMERGE_VALUES %3
S_ENDPGM 0, implicit %4, implicit %5, implicit %6, implicit %7
...
---
name: test_unmerge_values_s8_of_trunc_v4s16_concat_vectors_v2s32_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s8_of_trunc_v4s16_concat_vectors_v2s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
GlobalISel: Implement fewerElementsVector for G_TRUNC Extend fewerElementsVectorBasic to handle operands with different element types.
2020-02-16 09:24:36 +08:00
; CHECK: [[TRUNC:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[COPY]](<2 x s32>)
; CHECK: [[TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[COPY1]](<2 x s32>)
; CHECK: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8), [[UV2:%[0-9]+]]:_(s8), [[UV3:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[TRUNC]](<2 x s16>)
; CHECK: [[UV4:%[0-9]+]]:_(s8), [[UV5:%[0-9]+]]:_(s8), [[UV6:%[0-9]+]]:_(s8), [[UV7:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[TRUNC1]](<2 x s16>)
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
; CHECK: S_ENDPGM 0, implicit [[UV]](s8), implicit [[UV1]](s8), implicit [[UV2]](s8), implicit [[UV3]](s8), implicit [[UV4]](s8), implicit [[UV5]](s8), implicit [[UV6]](s8), implicit [[UV7]](s8)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<4 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s16>) = G_TRUNC %2
%4:_(s8), %5:_(s8), %6:_(s8), %7:_(s8), %8:_(s8), %9:_(s8), %10:_(s8), %11:_(s8) = G_UNMERGE_VALUES %3
S_ENDPGM 0, implicit %4, implicit %5, implicit %6, implicit %7, implicit %8, implicit %9, implicit %10, implicit %11
...
---
name: test_unmerge_values_s16_of_anyext_v4s64_concat_vectors_v2s32_v2s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s16_of_anyext_v4s64_concat_vectors_v2s32_v2s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
GlobalISel: Fold out unmerge to scalars from concat_vector Removes illegal intermediate vectors if an operation was lowering to concat_vectors, and the next operation is scalarized. llvm-svn: 367081
2019-07-26 10:22:23 +08:00
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[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: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[ANYEXT]](s64)
; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV2]](s32)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[UV2]], [[C]](s32)
; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; CHECK: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UV3]](s32)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[UV3]], [[C]](s32)
; CHECK: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; CHECK: S_ENDPGM 0, implicit [[TRUNC]](s16), implicit [[TRUNC1]](s16), implicit [[TRUNC2]](s16), implicit [[TRUNC3]](s16)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<4 x s32>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s64>) = G_ANYEXT %2
%4:_(s16), %5:_(s16), %6:_(s16), %7:_(s16), %8:_(s16), %9:_(s16), %10:_(s16), %11:_(s16), %12:_(s16), %13:_(s16), %14:_(s16), %15:_(s16), %16:_(s16), %17:_(s16), %18:_(s16), %19:_(s16) = G_UNMERGE_VALUES %3
S_ENDPGM 0, implicit %4, implicit %5, implicit %6, implicit %7
...
# FIXME: Handle this
---
name: test_unmerge_values_s32_of_concat_vectors_v4s32_v4s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s32_of_concat_vectors_v4s32_v4s32
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
; CHECK: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
; CHECK: [[COPY3:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr6_vgpr7
GlobalISel: Implement fewerElementsVector for G_TRUNC Extend fewerElementsVectorBasic to handle operands with different element types.
2020-02-16 09:24:36 +08:00
; CHECK: [[TRUNC:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[COPY]](<2 x s32>)
; CHECK: [[TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[COPY1]](<2 x s32>)
; CHECK: [[TRUNC2:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[COPY2]](<2 x s32>)
; CHECK: [[TRUNC3:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[COPY3]](<2 x s32>)
; CHECK: [[CONCAT_VECTORS:%[0-9]+]]:_(<8 x s16>) = G_CONCAT_VECTORS [[TRUNC]](<2 x s16>), [[TRUNC1]](<2 x s16>), [[TRUNC2]](<2 x s16>), [[TRUNC3]](<2 x s16>)
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<8 x s16>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: S_ENDPGM 0, implicit [[UV]](s32), implicit [[UV1]](s32), implicit [[UV2]](s32), implicit [[UV3]](s32)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<2 x s32>) = COPY $vgpr4_vgpr5
%3:_(<2 x s32>) = COPY $vgpr6_vgpr7
%4:_(<8 x s32>) = G_CONCAT_VECTORS %0, %1, %2, %3
%5:_(<8 x s16>) = G_TRUNC %4
%6:_(s32), %7:_(s32), %8:_(s32), %9:_(s32) = G_UNMERGE_VALUES %5
S_ENDPGM 0, implicit %6, implicit %7, implicit %8, implicit %9
...
---
name: test_unmerge_values_s64_of_build_vector_v4s32
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s64_of_build_vector_v4s32
; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
GlobalISel: Implement fewerElementsVector for G_TRUNC Extend fewerElementsVectorBasic to handle operands with different element types.
2020-02-16 09:24:36 +08:00
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32)
; CHECK: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[COPY2]](s32), [[COPY3]](s32)
; CHECK: [[TRUNC:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[BUILD_VECTOR]](<2 x s32>)
; CHECK: [[TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_TRUNC [[BUILD_VECTOR1]](<2 x s32>)
; CHECK: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[TRUNC]](<2 x s16>), [[TRUNC1]](<2 x s16>)
; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<4 x s16>)
GlobalISel: Combine unmerge of merge with intermediate cast This eliminates some illegal intermediate vectors when operations are scalarized. llvm-svn: 365566
2019-07-10 06:19:13 +08:00
; CHECK: S_ENDPGM 0, implicit [[UV]](s32), implicit [[UV1]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s32) = COPY $vgpr2
%3:_(s32) = COPY $vgpr3
%4:_(<4 x s32>) = G_BUILD_VECTOR %0, %1, %2, %3
%5:_(<4 x s16>) = G_TRUNC %4
%6:_(s32), %7:_(s32) = G_UNMERGE_VALUES %5
S_ENDPGM 0, implicit %6, implicit %7
...
[GISel][ArtifactCombiner] Relax the constraint to combine unmerge with concat_vectors The combine G_UNMERGE_VALUES with G_CONCAT_VECTORS used to only be performed when the result type of the G_UNMERGE_VALUES was a vector type. In other words, we were expecting that the G_UNMERGE_VALUES was effectively the exact opposite of the G_CONCAT_VECTORS. Lift that constraint by allowing any G_UNMERGE_VALUES to be combined with any G_CONCAT_VECTORS (as long as the size of the different pieces that we merge/unmerge match). Differential Revision: https://reviews.llvm.org/D69288
2019-11-07 03:09:12 +08:00
# To properly simplify that one, we would need to insert bitcast
# after the G_ZEXT.
# i.e.,
# s64 = zext <2 x s16> <-- invalid
# vs.
# <2 x s32> = zext <2 x s16>
# s64 = bitcast <2 x s32> <-- we are missing the code to do that
---
name: test_unmerge_values_s128_of_zext_of_concat_vectors
body: |
bb.0:
; CHECK-LABEL: name: test_unmerge_values_s128_of_zext_of_concat_vectors
; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
; 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: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
; CHECK: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C1]]
; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C1]]
; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C1]]
; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
; CHECK: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C1]]
; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[AND]](s32), [[AND1]](s32), [[AND2]](s32), [[AND3]](s32)
; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[BUILD_VECTOR]](<4 x s32>)
; CHECK: S_ENDPGM 0, implicit [[UV]](s64), implicit [[UV1]](s64)
%0:_(<2 x s16>) = COPY $vgpr0
%1:_(<2 x s16>) = COPY $vgpr1
%2:_(<4 x s16>) = G_CONCAT_VECTORS %0, %1
%3:_(<4 x s32>) = G_ZEXT %2
%4:_(s64), %5:_(s64) = G_UNMERGE_VALUES %3
S_ENDPGM 0, implicit %4, implicit %5
...